stable_datasets.images package

Submodules

stable_datasets.images.aiderv2 module

AIDERv2: aerial images of natural disasters (earthquake, flood, wildfire) and normal scenes.

class AIDERv2(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

AIDERv2: 16,723 aerial images in 4 classes, with official train/validation/test splits.

An extension of AIDER (Aerial Image Dataset for Emergency Response Applications). Labels are integers in [0, 4): 0 = Earthquake (collapsed buildings), 1 = Flood, 2 = Normal (no disaster), 3 = Wildfire. Splits: train (13,399), validation (1,670) and test (1,654). Images are returned as RGB at their original sizes.

Note: the images were gathered from several open datasets and from frames of YouTube videos, so the CC BY 4.0 license is the depositors’ label and the underlying sources have their own terms. Frames from the same video may appear in more than one split.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://zenodo.org/records/10891054', assets=mappingproxy({'train': DownloadInfo(url='https://zenodo.org/records/10891054/files/Train.zip?download=1', fallbacks=[], checksum=None, filename=None), 'validation': DownloadInfo(url='https://zenodo.org/records/10891054/files/Val.zip?download=1', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://zenodo.org/records/10891054/files/Test.zip?download=1', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{shianios2023benchmark,\n                      title={A Benchmark and Investigation of Deep-Learning-Based Techniques for Detecting Natural Disasters in Aerial Images},\n                      author={Shianios, D. and Kyrkou, C. and Kolios, P. S.},\n                      booktitle={Computer Analysis of Images and Patterns (CAIP)},\n                      series={Lecture Notes in Computer Science},\n                      volume={14185},\n                      publisher={Springer},\n                      year={2023},\n                      doi={10.1007/978-3-031-44240-7_24}}\n                    @article{shianios2024direcnetv2,\n                      title={DiRecNetV2: A Transformer-Enhanced Network for Aerial Disaster Recognition},\n                      author={Shianios, D. and Kolios, P. and Kyrkou, C.},\n                      journal={SN Computer Science},\n                      year={2024}}', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.arabic_characters module

class ArabicCharacters(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Arabic Handwritten Characters Dataset

Abstract Handwritten Arabic character recognition systems face several challenges, including the unlimited variation in human handwriting and large public databases. In this work, we model a deep learning architecture that can be effectively apply to recognizing Arabic handwritten characters. A Convolutional Neural Network (CNN) is a special type of feed-forward multilayer trained in supervised mode. The CNN trained and tested our database that contain 16800 of handwritten Arabic characters. In this paper, the optimization methods implemented to increase the performance of CNN. Common machine learning methods usually apply a combination of feature extractor and trainable classifier. The use of CNN leads to significant improvements across different machine-learning classification algorithms. Our proposed CNN is giving an average 5.1% misclassification error on testing data.

Context The motivation of this study is to use cross knowledge learned from multiple works to enhancement the performance of Arabic handwritten character recognition. In recent years, Arabic handwritten characters recognition with different handwriting styles as well, making it important to find and work on a new and advanced solution for handwriting recognition. A deep learning systems needs a huge number of data (images) to be able to make a good decisions.

Content The data-set is composed of 16,800 characters written by 60 participants, the age range is between 19 to 40 years, and 90% of participants are right-hand. Each participant wrote each character (from ’alef’ to ’yeh’) ten times on two forms as shown in Fig. 7(a) & 7(b). The forms were scanned at the resolution of 300 dpi. Each block is segmented automatically using Matlab 2016a to determining the coordinates for each block. The database is partitioned into two sets: a training set (13,440 characters to 480 images per class) and a test set (3,360 characters to 120 images per class). Writers of training set and test set are exclusive. Ordering of including writers to test set are randomized to make sure that writers of test set are not from a single institution (to ensure variability of the test set).

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/mloey/Arabic-Handwritten-Characters-Dataset', assets=mappingproxy({'train': DownloadInfo(url='https://github.com/mloey/Arabic-Handwritten-Characters-Dataset/raw/master/Train%20Images%2013440x32x32.zip', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://github.com/mloey/Arabic-Handwritten-Characters-Dataset/raw/master/Test%20Images%203360x32x32.zip', fallbacks=[], checksum=None, filename=None)}), citation='@article{el2017arabic,\n                        title={Arabic handwritten characters recognition using convolutional neural network},\n                        author={El-Sawy, Ahmed and Loey, Mohamed and El-Bakry, Hazem},\n                        journal={WSEAS Transactions on Computer Research},\n                        volume={5},\n                        pages={11--19},\n                        year={2017}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.arabic_digits module

class ArabicDigits(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Arabic Handwritten Digits Dataset.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/mloey/Arabic-Handwritten-Digits-Dataset', assets=mappingproxy({'train': DownloadInfo(url='https://raw.githubusercontent.com/mloey/Arabic-Handwritten-Digits-Dataset/master/Arabic%20Handwritten%20Digits%20Dataset%20CSV.zip', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://raw.githubusercontent.com/mloey/Arabic-Handwritten-Digits-Dataset/master/Arabic%20Handwritten%20Digits%20Dataset%20CSV.zip', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{el2016cnn,\n                        title={CNN for handwritten arabic digits recognition based on LeNet-5},\n                        author={El-Sawy, Ahmed and Hazem, EL-Bakry and Loey, Mohamed},\n                        booktitle={International conference on advanced intelligent systems and informatics},\n                        pages={566--575},\n                        year={2016},\n                        organization={Springer}\n                        }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.awa2 module

class AWA2(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

The Animals with Attributes 2 (AwA2) dataset provides images across 50 animal classes, useful for attribute-based classification and zero-shot learning research. See https://cvml.ista.ac.at/AwA2/ for more information.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://cvml.ista.ac.at/AwA2/', assets=mappingproxy({'train': DownloadInfo(url='https://cvml.ista.ac.at/AwA2/AwA2-data.zip', fallbacks=[], checksum=None, filename=None)}), citation='@ARTICLE{8413121,\n                         author={Xian, Yongqin and Lampert, Christoph H. and Schiele, Bernt and Akata, Zeynep},\n                         journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},\n                         title={Zero-Shot Learning—A Comprehensive Evaluation of the Good, the Bad and the Ugly},\n                         year={2019},\n                         volume={41},\n                         number={9},\n                         pages={2251-2265},\n                         keywords={Semantics;Visualization;Task analysis;Training;Fish;Protocols;Learning systems;Generalized zero-shot learning;transductive learning;image classification;weakly-supervised learning},\n                         doi={10.1109/TPAMI.2018.2857768}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.beans module

class Beans(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Bean disease dataset for classification of three classes: Angular Leaf Spot, Bean Rust, and Healthy leaves.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/AI-Lab-Makerere/ibean/', assets=mappingproxy({'train': DownloadInfo(url='https://storage.googleapis.com/ibeans/train.zip', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://storage.googleapis.com/ibeans/test.zip', fallbacks=[], checksum=None, filename=None), 'validation': DownloadInfo(url='https://storage.googleapis.com/ibeans/validation.zip', fallbacks=[], checksum=None, filename=None)}), citation='@misc{makerere2020beans,\n                         author = "{Makerere AI Lab}",\n                         title = "{Bean Disease Dataset}",\n                         year = "2020",\n                         month = "January",\n                         url = "https://github.com/AI-Lab-Makerere/ibean/"}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.bee_dataset module

BeeDataset: top-down photos of honey bees with multi-label annotations.

class BeeDataset(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

BeeDataset contains approximately 7,500 images with four binary labels.

Images show bees or wasps from above against a green background.

The labels are independent, so multiple labels may be active for one image: - varroa: varroa-mite infection - pollen: pollen packets are present - wasps: the image shows a wasp - cooling: cooling behavior is present

The dataset has no official train/test split, so all examples are exposed through a single train split.

Images in images_300 have a height of 300 and a width of 150.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/BeeAlarmed/BeeDataset', assets=mappingproxy({'train': DownloadInfo(url='https://github.com/BeeAlarmed/BeeDataset/raw/refs/heads/main/BeeDataset_20201121.zip', fallbacks=[], checksum=None, filename=None)}), citation='@misc{hickert2021beedataset,\n            title={Dataset for a camera based bee-hive monitoring},\n            author={Fabian Hickert},\n            year={2021},\n            url={https://github.com/BeeAlarmed/BeeDataset}\n        }', license='GPL-3.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.belgium_tsc module

class BelgiumTSC(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

BelgiumTSC: the Belgium Traffic Sign Classification benchmark.

Cropped traffic signs from the BelgiumTS database in 62 classes, following the GTSRB layout. Each image keeps a border of 10% of the sign size around the sign, and images vary in size. The official Training/Testing partitions are kept as released.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://btsd.ethz.ch/shareddata/', assets=mappingproxy({'train': DownloadInfo(url='https://btsd.ethz.ch/shareddata/BelgiumTSC/BelgiumTSC_Training.zip', fallbacks=[], checksum='sha256:e5f6a33c18716917d69989488b9f6af573821a7126ca1390af7e75403af1ee61', filename=None), 'test': DownloadInfo(url='https://btsd.ethz.ch/shareddata/BelgiumTSC/BelgiumTSC_Testing.zip', fallbacks=[], checksum='sha256:1811447452f02ede83c0237f2852fa8bd278f39e40ddcb1ba01c2ed212e735fa', filename=None)}), citation='@inproceedings{timofte2011sparse,\n                title     = {Sparse representation based projections},\n                author    = {Timofte, Radu and Van Gool, Luc},\n                booktitle = {British Machine Vision Conference},\n                year      = {2011} }\n                @article{timofte2014multi,\n                title     = {Multi-view traffic sign detection, recognition, and 3D localisation},\n                author    = {Timofte, Radu and Zimmermann, Karel and Van Gool, Luc},\n                journal   = {Machine Vision and Applications},\n                volume    = {25},\n                number    = {3},\n                pages     = {633--647},\n                year      = {2014} }\n        ', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.brain_tumor_mri module

class BrainTumorMRI(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Brain Tumor MRI Dataset

This dataset was put together by Masoud Nickparvar by combining three older datasets (Figshare, SARTAJ and Br35H). It has 7,200 brain MRI images, and each image belongs to one of four classes: glioma, meningioma, pituitary tumor or no tumor.

The training set has 1,400 images per class and the testing set has 400 per class. The images come in many different sizes and are a mix of grayscale and color, and they are kept exactly as published.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.kaggle.com/datasets/masoudnickparvar/brain-tumor-mri-dataset', assets=mappingproxy({'train': DownloadInfo(url='https://www.kaggle.com/api/v1/datasets/download/masoudnickparvar/brain-tumor-mri-dataset?datasetVersionNumber=2', fallbacks=[], checksum=None, filename='brain_tumor_mri.zip'), 'test': DownloadInfo(url='https://www.kaggle.com/api/v1/datasets/download/masoudnickparvar/brain-tumor-mri-dataset?datasetVersionNumber=2', fallbacks=[], checksum=None, filename='brain_tumor_mri.zip')}), citation='@misc{nickparvar2021braintumor,\n                      title={Brain Tumor MRI Dataset},\n                      author={Nickparvar, Masoud},\n                      year={2021},\n                      publisher={Kaggle},\n                      howpublished={\\url{https://www.kaggle.com/datasets/masoudnickparvar/brain-tumor-mri-dataset}}}', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.breakhis module

class BreakHis(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Breast Cancer Histopathological Image Classification (BreakHis).

BreakHis contains 7,909 microscopic images of breast tumour tissue from surgical biopsy slides of 82 patients, stained with hematoxylin and eosin and imaged at four magnifications (40X, 100X, 200X, 400X). The images are RGB PNGs of 700x460 pixels, except 74 that are 700x456. Each image is labelled with one of eight tumour types, four benign (adenosis, fibroadenoma, phyllodes tumour, tubular adenoma) and four malignant (ductal, lobular, mucinous and papillary carcinoma).

The authors publish the five train/test folds used in the paper, each of which keeps all images of a slide (one patient ID and tumour type) in the same split. Select one with config_name="fold1" (the default) to "fold5". Fold 1 has 5,005 train and 2,904 test images.

BUILDER_CONFIGS: list = [BuilderConfig(name='fold1', version=None, description='Official fold 1'), BuilderConfig(name='fold2', version=None, description='Official fold 2'), BuilderConfig(name='fold3', version=None, description='Official fold 3'), BuilderConfig(name='fold4', version=None, description='Official fold 4'), BuilderConfig(name='fold5', version=None, description='Official fold 5')]

DEFAULT_CONFIG_NAME: str | None = 'fold1'

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://web.inf.ufpr.br/vri/databases/breast-cancer-histopathological-database-breakhis/', assets=mappingproxy({'images': DownloadInfo(url='https://www.inf.ufpr.br/vri/databases/BreaKHis_v1.tar.gz', fallbacks=[], checksum='sha256:d118014cf8a58f6d1fc4571e55efb523c932c85cfba2fb05627e455fc40bd640', filename=None), 'folds': DownloadInfo(url='https://www.inf.ufpr.br/lesoliveira/download/mkfold.tar.gz', fallbacks=[], checksum='sha256:db7e2f16c398fbc42286b4f7d65e5f55c8f94d95123a46dfd052d8117d68ae27', filename=None)}), citation='@article{spanhol2016dataset,\n            title={A Dataset for Breast Cancer Histopathological Image Classification},\n            author={Spanhol, Fabio A and Oliveira, Luiz S and Petitjean, Caroline and Heutte, Laurent},\n            journal={IEEE Transactions on Biomedical Engineering},\n            volume={63},\n            number={7},\n            pages={1455--1462},\n            year={2016},\n            publisher={IEEE}\n        }', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.bsds500 module

class BSDS500(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Unsupervised image segmentation and boundary detection. The Berkeley Segmentation Dataset 500 contains 500 natural images (481x321 or 321x481), each with 4-9 human segmentations given as region-label maps and boundary maps. The official split is 200 train / 100 validation / 200 test images. No class labels.

Segmentations and boundaries are stored as lists of PNG bytes (one per annotator); use decode_mask to turn each into a 2-D integer array.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/resources.html', assets=mappingproxy({'train': DownloadInfo(url='http://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/BSR/BSR_bsds500.tgz', fallbacks=[], checksum=None, filename=None), 'validation': DownloadInfo(url='http://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/BSR/BSR_bsds500.tgz', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='http://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/BSR/BSR_bsds500.tgz', fallbacks=[], checksum=None, filename=None)}), citation="@article{arbelaez2011contour,\n                         title={Contour Detection and Hierarchical Image Segmentation},\n                         author={Arbel{\\'a}ez, Pablo and Maire, Michael and Fowlkes, Charless and Malik, Jitendra},\n                         journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},\n                         volume={33}, number={5}, pages={898--916}, year={2011}}", license='', checksums=None)

VERSION: Version = Version('1.0.0')

decode_mask(data: bytes) → ndarray

Decode PNG bytes produced by encode_mask back to a 2-D uint16 array.

encode_mask(mask: ndarray) → bytes

Encode a 2-D integer map as lossless 16-bit PNG bytes.

stable_datasets.images.caltech101 module

class Caltech101(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Caltech-101 object recognition dataset.

Caltech-101 contains pictures of objects belonging to 101 categories, collected by Fei-Fei Li, Marco Andreetto, and Marc’Aurelio Ranzato in 2003. Most categories have about 40 to 800 images, and most images are roughly 300x200 pixels. The archive also ships a BACKGROUND_Google clutter category, which is excluded here, as in torchvision, leaving 8677 images in 101 classes.

The dataset has no official train/test split, so this builder defines its own: for every class, file names are sorted, shuffled with a fixed seed (SPLIT_SEED = 0), and the first 30 images go to train while the remainder go to test. This gives 3030 training and 5647 test images. Grayscale images are converted to RGB.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://data.caltech.edu/records/mzrjq-6wc02', assets=mappingproxy({'train': DownloadInfo(url='https://data.caltech.edu/records/mzrjq-6wc02/files/caltech-101.zip?download=1', fallbacks=[], checksum='sha256:331234750fc7f77520e50d9565e8b6907b03565e30320da1401130db08c61f91', filename='caltech-101.zip'), 'test': DownloadInfo(url='https://data.caltech.edu/records/mzrjq-6wc02/files/caltech-101.zip?download=1', fallbacks=[], checksum='sha256:331234750fc7f77520e50d9565e8b6907b03565e30320da1401130db08c61f91', filename='caltech-101.zip')}), citation='@article{fei2006one,\n                    title     = {One-shot learning of object categories},\n                    author    = {Fei-Fei, Li and Fergus, Rob and Perona, Pietro},\n                    journal   = {IEEE Transactions on Pattern Analysis and Machine Intelligence},\n                    volume    = {28},\n                    number    = {4},\n                    pages     = {594--611},\n                    year      = {2006},\n                    doi       = {10.1109/TPAMI.2006.79}}', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.caltech101_silhouettes module

class Caltech101Silhouettes(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

CalTech 101 Silhouettes (28x28) with the official train/validation/test split.

Each image is the high-quality polygon outline of the primary object in a Caltech-101 image, centered, scaled and rendered as a filled black polygon on a white background. The split is stratified: each class contributes min(3/5 * N_c, 100) training images, and the remainder is divided evenly between validation and test, so the evaluation splits are class-imbalanced.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://people.cs.umass.edu/~marlin/data.shtml', assets=mappingproxy({'train': DownloadInfo(url='https://people.cs.umass.edu/~marlin/data/caltech101_silhouettes_28_split1.mat', fallbacks=[], checksum=None, filename=None), 'val': DownloadInfo(url='https://people.cs.umass.edu/~marlin/data/caltech101_silhouettes_28_split1.mat', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://people.cs.umass.edu/~marlin/data/caltech101_silhouettes_28_split1.mat', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{marlin2010inductive,\n                        title={Inductive Principles for Restricted Boltzmann Machine Learning},\n                        author={Marlin, Benjamin and Swersky, Kevin and Chen, Bo and de Freitas, Nando},\n                        booktitle={Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics},\n                        pages={509--516},\n                        year={2010},\n                        volume={9},\n                        series={Proceedings of Machine Learning Research},\n                        publisher={PMLR}}', license='Unknown', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.caltech256 module

class Caltech256(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Caltech-256 A collection of 30,607 training images with 256 categories and one additional clutter category (for testing background rejection) created by Griffin, Holub, and Perona (2007). Each category has a minimum of 80 images. There are 30,185 RGB and 422 grayscale images, all of which have been converted to contain 3 color channels, but the image sizes remain inconsistent.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://data.caltech.edu/records/nyy15-4j048', assets=mappingproxy({'train': DownloadInfo(url='https://data.caltech.edu/records/nyy15-4j048/files/256_ObjectCategories.tar?download=1', fallbacks=[], checksum='md5:67b4f42ca05d46448c6bb8ecd2220f6d', filename='256_ObjectCategories.tar')}), citation='@article{griffin_holub_perona_2007,\n                    title     = {Caltech-256 Object Category Dataset},\n                    publisher = {California Institute of Technology},\n                    author    = {Griffin, Gregory and Holub, Alex and Perona, Pietro},\n                    year      = {2007},\n                    month     = {mar}}', license='Creative Commons Attribution 4.0 International', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.cars196 module

class Cars196(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Cars-196 Dataset The Cars-196 dataset, also known as the Stanford Cars dataset, is a benchmark dataset for fine-grained visual classification of automobiles. It contains 16,185 color images covering 196 car categories, where each category is defined by a specific combination of make, model, and year. The dataset is split into 8,144 training images and 8,041 test images, with the first 98 classes used exclusively for training and the remaining 98 classes reserved for testing, ensuring that training and test classes are disjoint. Images are collected from real-world scenes and exhibit significant variation in v iewpoint, background, and lighting conditions. Each image is annotated with a class label and a tight bounding box around the car, making the dataset suitable for fine-grained recognition tasks that require precise object localization and strong generalization to unseen categories.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://ai.stanford.edu/~jkrause/cars/car_dataset.html', assets=mappingproxy({'train': DownloadInfo(url='https://huggingface.co/datasets/haodoz0118/cars196-img/resolve/main/cars196_train.zip', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://huggingface.co/datasets/haodoz0118/cars196-img/resolve/main/cars196_test.zip', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{krause20133d,\n            title={3d object representations for fine-grained categorization},\n            author={Krause, Jonathan and Stark, Michael and Deng, Jia and Fei-Fei, Li},\n            booktitle={Proceedings of the IEEE international conference on computer vision workshops},\n            pages={554--561},\n            year={2013}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.cars3d module

class Cars3D

Bases: BaseDatasetBuilder

183 car types x 24 azimuth angles x 4 elevation angles.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/google-research/disentanglement_lib/tree/master', assets=mappingproxy({'train': DownloadInfo(url='http://www.scottreed.info/files/nips2015-analogy-data.tar.gz', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{locatello2019challenging,\n  title={Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations},\n  author={Locatello, Francesco and Bauer, Stefan and Lucic, Mario and Raetsch, Gunnar and Gelly, Sylvain and Sch{"o}lkopf, Bernhard and Bachem, Olivier},\n  booktitle={International Conference on Machine Learning},\n  pages={4114--4124},\n  year={2019}\n}', license='Apache-2.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.cassava module

Cassava leaf disease image classification loader.

class cassava

Bases: object

Plant images classification.

The data consists of two folders, a training folder that contains 5 subfolders that contain the respective images for the different 5 classes and a test folder containing test images.

classes = ['cbb', 'cmd', 'cbsd', 'cgm', 'healthy']

static download(path)

Download cassavaleafdata.zip into <path>cassava/ unless it is already there.

static load(path=None)

Download if needed and load all splits as numpy arrays.

Parameters:

path – Folder holding the cassava/ data. Defaults to $DATASET_PATH.

Returns:

(train_images, train_labels, valid_images, valid_labels, test_images, test_labels).

stable_datasets.images.cct20 module

class CCT20(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Caltech Camera Traps (CCT20): camera-trap images of 15 animal and vehicle classes from 20 locations.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://lila.science/datasets/caltech-camera-traps', assets=mappingproxy({'images': DownloadInfo(url='https://storage.googleapis.com/public-datasets-lila/caltechcameratraps/eccv_18_all_images_sm.tar.gz', fallbacks=['https://lilawildlife.blob.core.windows.net/lila-wildlife/caltechcameratraps/eccv_18_all_images_sm.tar.gz'], checksum='md5:8143c17aa2a12872b66f284ff211531f', filename='cct20_images_sm.tar.gz'), 'annotations': DownloadInfo(url='https://storage.googleapis.com/public-datasets-lila/caltechcameratraps/eccv_18_annotations.tar.gz', fallbacks=['https://lilawildlife.blob.core.windows.net/lila-wildlife/caltechcameratraps/eccv_18_annotations.tar.gz'], checksum='md5:66a1f481b44aa1edadf75c9cfbd27aba', filename='cct20_annotations.tar.gz')}), citation='@inproceedings{beery2018recognition,\n                        title = {Recognition in Terra Incognita},\n                        author = {Beery, Sara and Van Horn, Grant and Perona, Pietro},\n                        booktitle = {Proceedings of the European Conference on Computer Vision (ECCV)},\n                        year = {2018}\n                        }', license='CDLA-Permissive-1.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.celeb_a module

class CelebA(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

The CelebA dataset is a large-scale face attributes dataset with more than 200K celebrity images, each with 40 attribute annotations.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html', assets=mappingproxy({'archive': DownloadInfo(url='https://drive.google.com/uc?export=download&id=0B7EVK8r0v71pZjFTYXZWM3FlRnM', fallbacks=[], checksum=None, filename=None), 'attributes': DownloadInfo(url='https://drive.google.com/uc?export=download&id=0B7EVK8r0v71pblRyaVFSWGxPY0U', fallbacks=[], checksum=None, filename=None), 'partition': DownloadInfo(url='https://drive.google.com/uc?export=download&id=0B7EVK8r0v71pY0NSMzRuSXJEVkk', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{liu2015faceattributes,\n                         title = {Deep Learning Face Attributes in the Wild},\n                         author = {Liu, Ziwei and Luo, Ping and Wang, Xiaogang and Tang, Xiaoou},\n                         booktitle = {Proceedings of International Conference on Computer Vision (ICCV)},\n                         month = {December},\n                         year = {2015}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.cifar10 module

class CIFAR10(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Image classification. The CIFAR-10 dataset was collected by Alex Krizhevsky, Vinod Nair, and Geoffrey Hinton. It consists of 60000 32x32 colour images in 10 classes, with 6000 images per class. There are 50000 training images and 10000 test images. The dataset is divided into five training batches and one test batch, each with 10000 images. The test batch contains exactly 1000 randomly selected images from each class. The training batches contain the remaining images in random order, but some training batches may contain more images from one class than another. Between them, the training batches contain exactly 5000 images from each class.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.cs.toronto.edu/~kriz/cifar.html', assets=mappingproxy({'train': DownloadInfo(url='https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz', fallbacks=[], checksum=None, filename=None)}), citation='@article{krizhevsky2009learning,\n                         title={Learning multiple layers of features from tiny images},\n                         author={Krizhevsky, Alex and Hinton, Geoffrey and others},\n                         year={2009},\n                         publisher={Toronto, ON, Canada}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.cifar100 module

class CIFAR100(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

CIFAR-100 dataset, a variant of CIFAR-10 with 100 classes.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.cs.toronto.edu/~kriz/cifar.html', assets=mappingproxy({'train': DownloadInfo(url='https://www.cs.toronto.edu/~kriz/cifar-100-python.tar.gz', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://www.cs.toronto.edu/~kriz/cifar-100-python.tar.gz', fallbacks=[], checksum=None, filename=None)}), citation='@article{krizhevsky2009learning,\n                         title={Learning multiple layers of features from tiny images},\n                         author={Krizhevsky, Alex and Hinton, Geoffrey and others},\n                         year={2009},\n                         publisher={Toronto, ON, Canada}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.cifar100_c module

class CIFAR100C(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

CIFAR-100-C dataset with corrupted CIFAR-100 images.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://zenodo.org/records/3555552', assets=mappingproxy({'test': DownloadInfo(url='https://zenodo.org/records/3555552/files/CIFAR-100-C.tar?download=1', fallbacks=[], checksum=None, filename=None)}), citation='@article{hendrycks2019robustness,\n                        title={Benchmarking Neural Network Robustness to Common Corruptions and Perturbations},\n                        author={Dan Hendrycks and Thomas Dietterich},\n                        journal={Proceedings of the International Conference on Learning Representations},\n                        year={2019}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.cifar100_n module

class CIFAR100N(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

CIFAR-100 training images with human-annotated CIFAR-100N labels.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/UCSC-REAL/cifar-10-100n', assets=mappingproxy({'cifar': DownloadInfo(url='https://www.cs.toronto.edu/~kriz/cifar-100-python.tar.gz', fallbacks=[], checksum=None, filename=None), 'annotations': DownloadInfo(url='https://raw.githubusercontent.com/UCSC-REAL/cifar-10-100n/main/data/CIFAR-100_human.pt', fallbacks=[], checksum=None, filename='CIFAR-100_human.pt')}), citation='\n@inproceedings{wei2022learning,\n  title={Learning with Noisy Labels Revisited: A Study Using Real-World Human\n  Annotations},\n  author={Jiaheng Wei and Zhaowei Zhu and Hao Cheng and Tongliang Liu and Gang\n  Niu and Yang Liu},\n  booktitle={International Conference on Learning Representations},\n  year={2022},\n  url={https://openreview.net/forum?id=TBWA6PLJZQm}\n}\n', license='CC BY-NC 4.0', checksums=None)

VERSION: Version = Version('1.0.1')

stable_datasets.images.cifar10_1 module

CIFAR-10.1, a fresh test set for measuring CIFAR-10 generalization.

class CIFAR10_1(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

The class-balanced v6 release of the CIFAR-10.1 test set.

CIFAR-10.1 contains 2,000 independently collected 32x32 RGB images. It uses the original CIFAR-10 label space and is intended for evaluating a model trained on CIFAR-10; it has no training split.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/modestyachts/CIFAR-10.1', assets=mappingproxy({'images': DownloadInfo(url='https://raw.githubusercontent.com/modestyachts/CIFAR-10.1/master/datasets/cifar10.1_v6_data.npy', fallbacks=[], checksum='sha256:2997188e5816f5bd545dc77771b6227828c28146049fcecf3fa10775474cacc6', filename='cifar10.1_v6_data.npy'), 'labels': DownloadInfo(url='https://raw.githubusercontent.com/modestyachts/CIFAR-10.1/master/datasets/cifar10.1_v6_labels.npy', fallbacks=[], checksum='sha256:ae40beda001693674edc94d925ee8268cfe68905f8f9aff800c8dcdfcd6c9448', filename='cifar10.1_v6_labels.npy')}), citation='@article{recht2018cifar10,\n  author={Recht, Benjamin and Roelofs, Rebecca and Schmidt, Ludwig and Shankar, Vaishaal},\n  title={Do CIFAR-10 Classifiers Generalize to CIFAR-10?},\n  journal={arXiv preprint arXiv:1806.00451},\n  year={2018}\n}', license='Unknown', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.cifar10_2 module

CIFAR-10.2, a reproduction-focused CIFAR-10 distribution-shift benchmark.

class CIFAR10_2(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

CIFAR-10.2 train and test splits for distribution-shift research.

The release reproduces the CIFAR-10 collection pipeline and provides 10,000 training examples plus a separate 2,000-example test set.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/modestyachts/cifar-10.2', assets=mappingproxy({'train': DownloadInfo(url='https://raw.githubusercontent.com/modestyachts/cifar-10.2/master/cifar102_train.npz', fallbacks=[], checksum='sha256:03804f609934c2fdf3adfb708a820e8372284ee5ccac1e4d8aed9f2c87cab099', filename='cifar102_train.npz'), 'test': DownloadInfo(url='https://raw.githubusercontent.com/modestyachts/cifar-10.2/master/cifar102_test.npz', fallbacks=[], checksum='sha256:e4fd6462bd5141ed293acd52ea570952f65d1bfd4ec9b068f87742f75a7a632e', filename='cifar102_test.npz')}), citation='@inproceedings{lu2020harder,\n  author={Lu, Shangyun and Nott, Bradley and Olson, Aaron and Todeschini, Alberto and\n          Vahabi, Hossein and Carmon, Yair and Schmidt, Ludwig},\n  title={Harder or Different? A Closer Look at Distribution Shift in Dataset Reproduction},\n  booktitle={ICML Workshop on Uncertainty and Robustness in Deep Learning},\n  year={2020},\n  url={https://www.gatsby.ucl.ac.uk/~balaji/udl2020/accepted-papers/UDL2020-paper-101.pdf}\n}', license='Unknown', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.cifar10_c module

class CIFAR10C(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

CIFAR-10-C dataset with corrupted CIFAR-10 images.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://zenodo.org/records/2535967', assets=mappingproxy({'test': DownloadInfo(url='https://zenodo.org/records/2535967/files/CIFAR-10-C.tar?download=1', fallbacks=[], checksum=None, filename=None)}), citation='@article{hendrycks2019robustness,\n                        title={Benchmarking Neural Network Robustness to Common Corruptions and Perturbations},\n                        author={Dan Hendrycks and Thomas Dietterich},\n                        journal={Proceedings of the International Conference on Learning Representations},\n                        year={2019}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.cifar10_n module

class CIFAR10N(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

CIFAR-10 training images with human-annotated CIFAR-10N labels.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/UCSC-REAL/cifar-10-100n', assets=mappingproxy({'cifar': DownloadInfo(url='https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz', fallbacks=[], checksum=None, filename=None), 'annotations': DownloadInfo(url='https://raw.githubusercontent.com/UCSC-REAL/cifar-10-100n/main/data/CIFAR-10_human.pt', fallbacks=[], checksum=None, filename='CIFAR-10_human.pt')}), citation='\n@inproceedings{wei2022learning,\n  title={Learning with Noisy Labels Revisited: A Study Using Real-World Human\n  Annotations},\n  author={Jiaheng Wei and Zhaowei Zhu and Hao Cheng and Tongliang Liu and Gang\n  Niu and Yang Liu},\n  booktitle={International Conference on Learning Representations},\n  year={2022},\n  url={https://openreview.net/forum?id=TBWA6PLJZQm}\n}\n', license='CC BY-NC 4.0', checksums=None)

VERSION: Version = Version('1.0.1')

stable_datasets.images.cinic10 module

class CINIC10(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

CINIC-10: CINIC-10 Is Not ImageNet or CIFAR-10.

CINIC-10 (Darlow et al., 2018) extends CIFAR-10 to 270,000 32x32 RGB images over the same ten classes: all 60,000 CIFAR-10 images plus 210,000 down-sampled ImageNet images. Each split (train, validation, test) has 90,000 class-balanced images, 20,000 of them from CIFAR-10. origin records which source each image comes from.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://datashare.ed.ac.uk/handle/10283/3192', assets=mappingproxy({'train': DownloadInfo(url='https://datashare.ed.ac.uk/server/api/core/bitstreams/e8e186cc-2688-48f1-aa5a-3fda1a43f8b6/content', fallbacks=['https://datashare.ed.ac.uk/bitstreams/e8e186cc-2688-48f1-aa5a-3fda1a43f8b6/download'], checksum='md5:6ee4d0c996905fe93221de577967a372', filename='CINIC-10.tar.gz'), 'validation': DownloadInfo(url='https://datashare.ed.ac.uk/server/api/core/bitstreams/e8e186cc-2688-48f1-aa5a-3fda1a43f8b6/content', fallbacks=['https://datashare.ed.ac.uk/bitstreams/e8e186cc-2688-48f1-aa5a-3fda1a43f8b6/download'], checksum='md5:6ee4d0c996905fe93221de577967a372', filename='CINIC-10.tar.gz'), 'test': DownloadInfo(url='https://datashare.ed.ac.uk/server/api/core/bitstreams/e8e186cc-2688-48f1-aa5a-3fda1a43f8b6/content', fallbacks=['https://datashare.ed.ac.uk/bitstreams/e8e186cc-2688-48f1-aa5a-3fda1a43f8b6/download'], checksum='md5:6ee4d0c996905fe93221de577967a372', filename='CINIC-10.tar.gz')}), citation='@article{darlow2018cinic,\n                        title={CINIC-10 is not ImageNet or CIFAR-10},\n                        author={Darlow, Luke N and Crowley, Elliot J and Antoniou, Antreas and Storkey, Amos J},\n                        journal={arXiv preprint arXiv:1810.03505},\n                        year={2018}}', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.clevrer module

class CLEVRER(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

CLEVRER: CoLlision Events for Video REpresentation and Reasoning.

A diagnostic video dataset for systematic evaluation of computational models on a wide range of reasoning tasks. The dataset includes four types of questions: descriptive (e.g., “what color”), explanatory (“what’s responsible for”), predictive (“what will happen next”), and counterfactual (“what if”).

The dataset contains 20,000 synthetic videos of moving and colliding objects. Each video is 5 seconds long and contains 128 frames with resolution 480 x 320.

Splits:

• train: 10,000 videos (index 0 - 9999)

• validation: 5,000 videos (index 10000 - 14999)

• test: 5,000 videos (index 15000 - 19999)

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='http://clevrer.csail.mit.edu/', assets=mappingproxy({'train_videos': DownloadInfo(url='http://data.csail.mit.edu/clevrer/videos/train/video_train.zip', fallbacks=[], checksum=None, filename=None), 'train_annotations': DownloadInfo(url='http://data.csail.mit.edu/clevrer/annotations/train/annotation_train.zip', fallbacks=[], checksum=None, filename=None), 'train_questions': DownloadInfo(url='http://data.csail.mit.edu/clevrer/questions/train.json', fallbacks=[], checksum=None, filename=None), 'validation_videos': DownloadInfo(url='http://data.csail.mit.edu/clevrer/videos/validation/video_validation.zip', fallbacks=[], checksum=None, filename=None), 'validation_annotations': DownloadInfo(url='http://data.csail.mit.edu/clevrer/annotations/validation/annotation_validation.zip', fallbacks=[], checksum=None, filename=None), 'validation_questions': DownloadInfo(url='http://data.csail.mit.edu/clevrer/questions/validation.json', fallbacks=[], checksum=None, filename=None), 'test_videos': DownloadInfo(url='http://data.csail.mit.edu/clevrer/videos/test/video_test.zip', fallbacks=[], checksum=None, filename=None), 'test_questions': DownloadInfo(url='http://data.csail.mit.edu/clevrer/questions/test.json', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{yi2020clevrer,\n            title={CLEVRER: CoLlision Events for Video REpresentation and Reasoning},\n            author={Yi, Kexin and Gan, Chuang and Li, Yunzhu and Kohli, Pushmeet and Wu, Jiajun and Torralba, Antonio and Tenenbaum, Joshua B},\n            booktitle={International Conference on Learning Representations},\n            year={2020}\n        }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.cmu_faces module

CMU Face Images with a deterministic subject-disjoint split.

class CMUFaces(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Four-class head-pose classification using 32-by-30 grayscale images.

This builder defines a custom 12/4/4-person train/validation/test split. It is not an official benchmark split. Other image resolutions are excluded so that resized copies cannot enter different splits.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://archive.ics.uci.edu/dataset/124/cmu+face+images', assets=mappingproxy({'train': DownloadInfo(url='https://archive.ics.uci.edu/static/public/124/cmu+face+images.zip', fallbacks=[], checksum='sha256:01a8b6cc58ac2f825f72abd6cf2b393df067c6529d96f7a90b22dede71b142cc', filename=None), 'validation': DownloadInfo(url='https://archive.ics.uci.edu/static/public/124/cmu+face+images.zip', fallbacks=[], checksum='sha256:01a8b6cc58ac2f825f72abd6cf2b393df067c6529d96f7a90b22dede71b142cc', filename=None), 'test': DownloadInfo(url='https://archive.ics.uci.edu/static/public/124/cmu+face+images.zip', fallbacks=[], checksum='sha256:01a8b6cc58ac2f825f72abd6cf2b393df067c6529d96f7a90b22dede71b142cc', filename=None)}), citation='@misc{cmu_face_images_124,\n    author = {Mitchell, Tom},\n    title = {CMU Face Images},\n    publisher = {UCI Machine Learning Repository},\n    year = {1997},\n    doi = {10.24432/C5JC79},\n    url = {https://archive.ics.uci.edu/dataset/124/cmu+face+images}\n}', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.coil100 module

Columbia Object Image Library (COIL-100).

class COIL100(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

100 physical objects, each photographed at 72 viewing angles.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.cs.columbia.edu/CAVE/software/softlib/coil-100.php', assets=mappingproxy({'train': DownloadInfo(url='https://www.cs.columbia.edu/CAVE/databases/SLAM_coil-20_coil-100/coil-100/coil-100.zip', fallbacks=[], checksum=None, filename=None)}), citation='@techreport{nene1996coil100,\n  author = {Nene, Sameer A. and Nayar, Shree K. and Murase, Hiroshi},\n  title = {Columbia Object Image Library (COIL-100)},\n  institution = {Columbia University},\n  number = {CUCS-006-96},\n  month = feb,\n  year = {1996}\n}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.coil20 module

class COIL20(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

COIL-20 images of 20 objects photographed from multiple views.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.cs.columbia.edu/CAVE/software/softlib/coil-20.php', assets=mappingproxy({'train': DownloadInfo(url='https://www.cs.columbia.edu/CAVE/databases/SLAM_coil-20_coil-100/coil-20/coil-20-proc.zip', fallbacks=[], checksum=None, filename=None)}), citation='@techreport{nene1996coil20,\n            title={Columbia Object Image Library (COIL-20)},\n            author={Nene, Sameer A. and Nayar, Shree K. and Murase, Hiroshi},\n            institution={Columbia University},\n            year={1996}\n        }', license='Unspecified: no dataset-specific license identified in the official source. See the documentation for source links and usage information.', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.country211 module

class Country211(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Country211: Image Classification Dataset for Geolocation. This dataset uses a subset of the YFCC100M dataset, filtered by GPS coordinates to include images labeled with ISO-3166 country codes. Each country has a balanced sample of images for training, validation, and testing.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/openai/CLIP/blob/main/data/country211.md', assets=mappingproxy({'train': DownloadInfo(url='https://openaipublic.azureedge.net/clip/data/country211.tgz', fallbacks=[], checksum=None, filename=None), 'valid': DownloadInfo(url='https://openaipublic.azureedge.net/clip/data/country211.tgz', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://openaipublic.azureedge.net/clip/data/country211.tgz', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{radford2021learning,\n                title     = {Learning transferable visual models from natural language supervision},\n                author    = {Radford, Alec and Kim, Jong Wook and Hallacy, Chris and Ramesh, Aditya and Goh, Gabriel and Agarwal, Sandhini and Sastry, Girish and Askell, Amanda and Mishkin, Pamela and Clark, Jack and others},\n                booktitle = {International conference on machine learning},\n                pages     = {8748--8763},\n                year      = {2021},\n                organization = {PmLR} }\n        ', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.cub200 module

class CUB200(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Caltech-UCSD Birds-200-2011 (CUB-200-2011) Dataset

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.vision.caltech.edu/datasets/cub_200_2011/', assets=mappingproxy({'train': DownloadInfo(url='https://data.caltech.edu/records/65de6-vp158/files/CUB_200_2011.tgz?download=1', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://data.caltech.edu/records/65de6-vp158/files/CUB_200_2011.tgz?download=1', fallbacks=[], checksum=None, filename=None)}), citation='@techreport{WahCUB_200_2011,\n                        Title = {The Caltech-UCSD Birds-200-2011 Dataset},\n                        Author = {Wah, C. and Branson, S. and Welinder, P. and Perona, P. and Belongie, S.},\n                        Year = {2011},\n                        Institution = {California Institute of Technology},\n                        Number = {CNS-TR-2011-001}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.deep_weeds module

class DeepWeeds(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

DeepWeeds dataset for classification of eight classes of weeds as well as a negative class in the Australian rangelands.

BUILDER_CONFIGS: list = [BuilderConfig(name='subset0', version=None, description='Official train/val/test split 0 of 5'), BuilderConfig(name='subset1', version=None, description='Official train/val/test split 1 of 5'), BuilderConfig(name='subset2', version=None, description='Official train/val/test split 2 of 5'), BuilderConfig(name='subset3', version=None, description='Official train/val/test split 3 of 5'), BuilderConfig(name='subset4', version=None, description='Official train/val/test split 4 of 5')]

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/AlexOlsen/DeepWeeds', assets=mappingproxy({'images': DownloadInfo(url='https://drive.usercontent.google.com/download?id=1xnK3B6K6KekDI55vwJ0vnc2IGoDga9cj&export=download&confirm=t', fallbacks=[], checksum=None, filename='deepweeds_images.zip'), 'labels': DownloadInfo(url='https://codeload.github.com/AlexOlsen/DeepWeeds/zip/da084f62bcd1a2b0afeb8b81f1a27be3186391a1', fallbacks=[], checksum=None, filename='deepweeds_labels.zip')}), citation='@article{DeepWeeds2019,\n                        author = {Alex Olsen and\n                            Dmitry A. Konovalov and\n                            Bronson Philippa and\n                            Peter Ridd and\n                            Jake C. Wood and\n                            Jamie Johns and\n                            Wesley Banks and\n                            Benjamin Girgenti and\n                            Owen Kenny and\n                            James Whinney and\n                            Brendan Calvert and\n                            Mostafa {Rahimi Azghadi} and\n                            Ronald D. White},\n                        title = {{DeepWeeds: A Multiclass Weed Species Image Dataset for Deep Learning}},\n                        journal = {Scientific Reports},\n                        year = 2019,\n                        number = 2058,\n                        month = 2,\n                        volume = 9,\n                        issue = 1,\n                        day = 14,\n                        url = "https://doi.org/10.1038/s41598-018-38343-3",\n                        doi = "10.1038/s41598-018-38343-3"\n                        }\n                        ', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.defungi module

DeFungi images with deterministic, duplicate-linked source-group splits.

class DeFungi(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Load all 9,114 released RGB patches, preserving original class codes.

A custom split holds out filename-derived source groups and links groups containing byte-identical images. Source groups are not patient identities.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://archive.ics.uci.edu/dataset/773/defungi', assets=mappingproxy({'train': DownloadInfo(url='https://archive.ics.uci.edu/static/public/773/defungi.zip', fallbacks=[], checksum='sha256:02020311d1f8207f3aa3bf8b2303a5106463ac0bf35adde4ae042e8b2716ff6c', filename=None), 'validation': DownloadInfo(url='https://archive.ics.uci.edu/static/public/773/defungi.zip', fallbacks=[], checksum='sha256:02020311d1f8207f3aa3bf8b2303a5106463ac0bf35adde4ae042e8b2716ff6c', filename=None), 'test': DownloadInfo(url='https://archive.ics.uci.edu/static/public/773/defungi.zip', fallbacks=[], checksum='sha256:02020311d1f8207f3aa3bf8b2303a5106463ac0bf35adde4ae042e8b2716ff6c', filename=None)}), citation='@misc{sopo2021defungi,\n    title = {DeFungi: Direct Mycological Examination of Microscopic Fungi Images},\n    author = {Pineda Sopo, Camilo Javier and Hajati, Farshid and Gheisari, Soheila},\n    year = {2021},\n    eprint = {2109.07322},\n    archivePrefix = {arXiv},\n    doi = {10.48550/arXiv.2109.07322}\n}', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.devanagari module

class Devanagari(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

DHCD: 46 handwritten character classes with the original train/test split.

CLASS_NAMES = ('character_1_ka', 'character_2_kha', 'character_3_ga', 'character_4_gha', 'character_5_kna', 'character_6_cha', 'character_7_chha', 'character_8_ja', 'character_9_jha', 'character_10_yna', 'character_11_taamatar', 'character_12_thaa', 'character_13_daa', 'character_14_dhaa', 'character_15_adna', 'character_16_tabala', 'character_17_tha', 'character_18_da', 'character_19_dha', 'character_20_na', 'character_21_pa', 'character_22_pha', 'character_23_ba', 'character_24_bha', 'character_25_ma', 'character_26_yaw', 'character_27_ra', 'character_28_la', 'character_29_waw', 'character_30_motosaw', 'character_31_petchiryakha', 'character_32_patalosaw', 'character_33_ha', 'character_34_chhya', 'character_35_tra', 'character_36_gya', 'digit_0', 'digit_1', 'digit_2', 'digit_3', 'digit_4', 'digit_5', 'digit_6', 'digit_7', 'digit_8', 'digit_9')

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://archive.ics.uci.edu/dataset/389/devanagari+handwritten+character+dataset', assets=mappingproxy({'train': DownloadInfo(url='https://archive.ics.uci.edu/static/public/389/devanagari+handwritten+character+dataset.zip', fallbacks=[], checksum='sha256:752e6a2e6b0d1b475375e6a9db937918bbbf7c6b7d137dbe234264029fd3b88d', filename=None), 'test': DownloadInfo(url='https://archive.ics.uci.edu/static/public/389/devanagari+handwritten+character+dataset.zip', fallbacks=[], checksum='sha256:752e6a2e6b0d1b475375e6a9db937918bbbf7c6b7d137dbe234264029fd3b88d', filename=None)}), citation='@misc{dhcd2015,\n            author={Acharya, Shailesh and Gyawali, Prashnna},\n            title={Devanagari Handwritten Character Dataset},\n            year={2015}, publisher={UCI Machine Learning Repository},\n            doi={10.24432/C5XS53}}\n        ', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.dsprites module

class DSprites

Bases: BaseDatasetBuilder

dSprites is a dataset of 2D shapes procedurally generated from 6 ground truth independent latent factors. These factors are color, shape, scale, rotation, x and y positions of a sprite.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/deepmind/dsprites-dataset', assets=mappingproxy({'train': DownloadInfo(url='https://github.com/google-deepmind/dsprites-dataset/raw/refs/heads/master/dsprites_ndarray_co1sh3sc6or40x32y32_64x64.npz', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{higgins2017beta,\n                    title={beta-vae: Learning basic visual concepts with a constrained variational framework},\n                    author={Higgins, Irina and Matthey, Loic and Pal, Arka and Burgess, Christopher and Glorot, Xavier and Botvinick, Matthew and Mohamed, Shakir and Lerchner, Alexander},\n                    booktitle={International conference on learning representations},\n                    year={2017}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.dsprites_color module

class DSpritesColor(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

DSprites dSprites is a dataset of 2D shapes procedurally generated from 6 ground truth independent latent factors. These factors are color, shape, scale, rotation, x and y positions of a sprite.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/deepmind/dsprites-dataset', assets=mappingproxy({'train': DownloadInfo(url='https://github.com/google-deepmind/dsprites-dataset/raw/refs/heads/master/dsprites_ndarray_co1sh3sc6or40x32y32_64x64.npz', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{locatello2019challenging,\n                    title={Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations},\n                    author={Locatello, Francesco and Bauer, Stefan and Lucic, Mario and Raetsch, Gunnar and Gelly, Sylvain and Sch{"o}lkopf, Bernhard and Bachem, Olivier},\n                    booktitle={International Conference on Machine Learning},\n                    pages={4114--4124},\n                    year={2019}\n                    }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.dsprites_noise module

class DSpritesNoise(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

DSprites dSprites is a dataset of 2D shapes procedurally generated from 6 ground truth independent latent factors. These factors are color, shape, scale, rotation, x and y positions of a sprite.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/deepmind/dsprites-dataset', assets=mappingproxy({'train': DownloadInfo(url='https://github.com/google-deepmind/dsprites-dataset/raw/refs/heads/master/dsprites_ndarray_co1sh3sc6or40x32y32_64x64.npz', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{locatello2019challenging,\n                    title={Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations},\n                    author={Locatello, Francesco and Bauer, Stefan and Lucic, Mario and Raetsch, Gunnar and Gelly, Sylvain and Sch{"o}lkopf, Bernhard and Bachem, Olivier},\n                    booktitle={International Conference on Machine Learning},\n                    pages={4114--4124},\n                    year={2019}\n                    }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.dsprites_scream module

class DSpritesScream(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

DSprites dSprites is a dataset of 2D shapes procedurally generated from 6 ground truth independent latent factors. These factors are color, shape, scale, rotation, x and y positions of a sprite.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/deepmind/dsprites-dataset', assets=mappingproxy({'train': DownloadInfo(url='https://github.com/google-deepmind/dsprites-dataset/raw/refs/heads/master/dsprites_ndarray_co1sh3sc6or40x32y32_64x64.npz', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{higgins2017beta,\n                    title={beta-vae: Learning basic visual concepts with a constrained variational framework},\n                    author={Higgins, Irina and Matthey, Loic and Pal, Arka and Burgess, Christopher and Glorot, Xavier and Botvinick, Matthew and Mohamed, Shakir and Lerchner, Alexander},\n                    booktitle={International conference on learning representations},\n                    year={2017}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.dtd module

class DTD(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Describable Textures Dataset (DTD)

DTD is a texture database, consisting of 5640 images, organized according to a list of 47 terms (categories) inspired from human perception. There are 120 images for each category. Image sizes range between 300x300 and 640x640, and the images contain at least 90% of the surface representing the category attribute. The images were collected from Google and Flickr by entering our proposed attributes and related terms as search queries. The images were annotated using Amazon Mechanical Turk in several iterations. For each image we provide key attribute (main category) and a list of joint attributes.

The data is split in three equal parts, in train, validation and test, 40 images per class, for each split. We provide the ground truth annotation for both key and joint attributes, as well as the 10 splits of the data we used for evaluation.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.robots.ox.ac.uk/~vgg/data/dtd/', assets=mappingproxy({'train': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/dtd/download/dtd-r1.0.1.tar.gz', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/dtd/download/dtd-r1.0.1.tar.gz', fallbacks=[], checksum=None, filename=None), 'val': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/dtd/download/dtd-r1.0.1.tar.gz', fallbacks=[], checksum=None, filename=None)}), citation='@InProceedings{cimpoi14describing,\n                    Author    = {M. Cimpoi and S. Maji and I. Kokkinos and S. Mohamed and and A. Vedaldi},\n                    Title     = {Describing Textures in the Wild},\n                    Booktitle = {Proceedings of the {IEEE} Conf. on Computer Vision and Pattern Recognition ({CVPR})},\n                    Year      = {2014}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.e_mnist module

class EMNIST(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

EMNIST (Extended MNIST) Dataset

Abstract EMNIST is a set of handwritten characters derived from the NIST Special Database 19 and converted to a 28x28 pixel format that directly matches the MNIST dataset. It serves as a challenging “drop-in” replacement for MNIST, introducing handwritten letters and a larger variety of writing styles while preserving the original file structure and pixel density.

Context While the original MNIST dataset is considered “solved” by modern architectures, EMNIST restores the challenge by providing a larger, more diverse benchmark. It bridges the gap between simple digit recognition and complex handwriting tasks, offering up to 62 classes (digits + uppercase + lowercase) to test generalization and writer-independent recognition.

Content The dataset contains up to 814,255 grayscale images (28x28). It is provided in six split configurations to suit different needs: * ByClass & ByMerge: Full unbalanced sets (up to 62 classes). * Balanced: 131,600 images across 47 classes (ideal for benchmarking). * Letters: 145,600 images across 26 classes (A-Z). * Digits & MNIST: 280,000+ images across 10 classes (0-9).

BUILDER_CONFIGS: list = [EMNISTConfig(name='byclass', version=Version('1.0.0'), description=''), EMNISTConfig(name='bymerge', version=Version('1.0.0'), description=''), EMNISTConfig(name='balanced', version=Version('1.0.0'), description=''), EMNISTConfig(name='letters', version=Version('1.0.0'), description=''), EMNISTConfig(name='digits', version=Version('1.0.0'), description=''), EMNISTConfig(name='mnist', version=Version('1.0.0'), description='')]

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.nist.gov/itl/iad/image-group/emnist-dataset', assets=mappingproxy({'train': DownloadInfo(url='https://biometrics.nist.gov/cs_links/EMNIST/matlab.zip', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://biometrics.nist.gov/cs_links/EMNIST/matlab.zip', fallbacks=[], checksum=None, filename=None)}), citation='@misc{cohen2017emnistextensionmnisthandwritten,\n                        title={EMNIST: an extension of MNIST to handwritten letters},\n                        author={Gregory Cohen and Saeed Afshar and Jonathan Tapson and André van Schaik},\n                        year={2017},\n                        eprint={1702.05373},\n                        archivePrefix={arXiv},\n                        primaryClass={cs.CV},\n                        url={https://arxiv.org/abs/1702.05373},\n            }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

class EMNISTConfig(variant, **kwargs)

Bases: BuilderConfig

Config for one EMNIST split (byclass, bymerge, balanced, letters, digits or mnist).

stable_datasets.images.eth80 module

ETH-80 cropped256 images with a fixed, instance-disjoint split.

class ETH80(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Eight categories, ten objects per category and 41 views per object.

This contribution defines the split (it is not an official benchmark): the first six objects train, the next two validation, and the last two test, ordered by numeric instance ID within each class. object_id is globally unique; instance_id is local to the class. Segmentation maps in the archive are excluded.

INSTANCE_IDS = {'apple': (1, 2, 3, 4, 5, 6, 7, 8, 9, 10), 'car': (1, 2, 3, 5, 6, 7, 9, 11, 12, 14), 'cow': (1, 2, 3, 4, 5, 6, 7, 8, 9, 10), 'cup': (1, 2, 3, 4, 5, 6, 7, 8, 9, 10), 'dog': (1, 2, 3, 4, 5, 6, 7, 8, 9, 10), 'horse': (1, 2, 3, 4, 5, 6, 7, 8, 9, 10), 'pear': (1, 2, 3, 4, 5, 6, 7, 8, 9, 10), 'tomato': (1, 2, 3, 4, 5, 6, 7, 8, 9, 10)}

LABELS = ('apple', 'car', 'cow', 'cup', 'dog', 'horse', 'pear', 'tomato')

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.mpi-inf.mpg.de/departments/computer-vision-and-machine-learning/research/object-recognition-and-scene-understanding/analyzing-appearance-and-contour-based-methods-for-object-categorization', assets=mappingproxy({'train': DownloadInfo(url='https://datasets.d2.mpi-inf.mpg.de/eth80/eth80-cropped256.tgz', fallbacks=[], checksum='sha256:3552a8199bbf011bcc795364496430dbf936816f89b96ddb7a28af67cc6e6c95', filename=None), 'validation': DownloadInfo(url='https://datasets.d2.mpi-inf.mpg.de/eth80/eth80-cropped256.tgz', fallbacks=[], checksum='sha256:3552a8199bbf011bcc795364496430dbf936816f89b96ddb7a28af67cc6e6c95', filename=None), 'test': DownloadInfo(url='https://datasets.d2.mpi-inf.mpg.de/eth80/eth80-cropped256.tgz', fallbacks=[], checksum='sha256:3552a8199bbf011bcc795364496430dbf936816f89b96ddb7a28af67cc6e6c95', filename=None)}), citation='@inproceedings{leibe2003analyzing,\n    title={Analyzing Appearance and Contour Based Methods for Object Categorization},\n    author={Leibe, Bastian and Schiele, Bernt},\n    booktitle={IEEE Conference on Computer Vision and Pattern Recognition},\n    year={2003}\n}', license='No explicit dataset license found on the provider page; consult the original provider for reuse terms.', checksums=None)

SPLIT_SLICES = {'test': slice(8, 10, None), 'train': slice(0, 6, None), 'validation': slice(6, 8, None)}

VERSION: Version = Version('1.0.0')

stable_datasets.images.eurosat module

class EuroSAT(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

EuroSAT Dataset

Abstract A land use / land cover classification dataset built from Sentinel-2 satellite imagery, covering 10 classes with 27,000 labeled 64x64 RGB images in total.

There is no official train/val/test split shipped with the archive (images are organized one folder per class), so this builder exposes a single “train” split with every image (the same pattern as AWA2).

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/phelber/EuroSAT', assets=mappingproxy({'train': DownloadInfo(url='https://huggingface.co/datasets/torchgeo/eurosat/resolve/main/EuroSAT.zip', fallbacks=['https://madm.dfki.de/files/sentinel/EuroSAT.zip'], checksum=None, filename=None)}), citation='@article{helber2019eurosat,\n                         title={EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification},\n                         author={Helber, Patrick and Bischke, Benjamin and Dengel, Andreas and Borth, Damian},\n                         journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},\n                         year={2019}}', license='MIT License', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.face_pointing module

class FacePointing(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Head angle classification dataset.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='http://crowley-coutaz.fr/HeadPoseDataSet/', assets=mappingproxy({'train': DownloadInfo(url='http://crowley-coutaz.fr/HeadPoseDataSet/HeadPoseImageDatabase.tar.gz', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{gourier2004estimating,\n                         title={Estimating face orientation from robust detection of salient facial features},\n                         author={Gourier, Nicolas and Hall, Daniela and Crowley, James L},\n                         booktitle={ICPR International Workshop on Visual Observation of Deictic Gestures},\n                         year={2004},\n                         organization={Citeseer}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.fashion_mnist module

class FashionMNIST(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Grayscale image classification.

Fashion-MNIST is a dataset of Zalando’s article images consisting of a training set of 60,000 examples and a test set of 10,000 examples. Each example is a 28x28 grayscale image, associated with a label from 10 classes.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/zalandoresearch/fashion-mnist', assets=mappingproxy({'train': DownloadInfo(url='http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-images-idx3-ubyte.gz', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-images-idx3-ubyte.gz', fallbacks=[], checksum=None, filename=None)}), citation='@article{xiao2017fashion,\n                         title={Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms},\n                         author={Xiao, Han and Rasul, Kashif and Vollgraf, Roland},\n                         journal={arXiv preprint arXiv:1708.07747},\n                         year={2017}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.fer2013 module

class FER2013(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

FER2013 (Facial Expression Recognition 2013).

Abstract 48x48 grayscale face crops labelled with one of seven facial expressions: angry, disgust, fear, happy, sad, surprise and neutral. Released for the ICML 2013 “Challenges in Representation Learning” Kaggle competition, it is one of the 27 datasets of CLIP’s linear-probe evaluation suite.

Context The faces were collected with Google image searches for emotion-related keywords, then automatically registered and cropped, so pose, lighting, age and occlusion vary widely. Human accuracy is estimated at ~65%, and the classes are imbalanced (disgust has 547 examples, happy 8,989), which makes it a demanding transfer benchmark.

Content

• Images: 35,887 grayscale 48x48 faces, decoded from the space-separated pixels column of the competition CSV.

• Labels: the emotion column, 0-6 in the order angry, disgust, fear, happy, sad, surprise, neutral.

• Splits: the competition’s Usage column. Training -> train (28,709), PublicTest -> validation (3,589), PrivateTest -> test (3,589).

Download Kaggle requires an account and acceptance of the competition rules, so the data cannot be fetched automatically. Download fer2013.csv (or fer2013.tar.gz / icml_face_data.csv) from the competition page and place it in the builder’s download directory (download_dir, default ~/.stable-datasets/downloads).

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.kaggle.com/c/challenges-in-representation-learning-facial-expression-recognition-challenge', assets=mappingproxy({}), citation='@inproceedings{goodfellow2013challenges,\n                    title={Challenges in Representation Learning: A report on three machine learning contests},\n                    author={Goodfellow, Ian J. and Erhan, Dumitru and Carrier, Pierre Luc and Courville, Aaron and\n                            Mirza, Mehdi and Hamner, Ben and Cukierski, Will and Tang, Yichuan and Thaler, David and\n                            Lee, Dong-Hyun and others},\n                    booktitle={International Conference on Neural Information Processing (ICONIP)},\n                    year={2013}}', license='Kaggle competition rules', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.fgvc_aircraft module

class FGVCAircraft(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Fine-Grained Visual Classification of Aircraft (FGVC-Aircraft) Dataset.

FGVC-Aircraft is a benchmark dataset for fine-grained visual categorization of aircraft. The dataset contains 10,000 images of aircraft with 100 different aircraft model variants. Aircraft models are organized in a hierarchical structure with three levels: variant (finest), family, and manufacturer (coarsest).

The dataset is divided into training (3,334 images), validation (3,333 images), and test (3,333 images) subsets. Images are about 1-2MP resolution with a 20-pixel copyright banner at the bottom that is automatically removed during loading.

Usage:

dataset = FGVCAircraft(config_name=”variant”, split=”train”) dataset = FGVCAircraft(config_name=”family”, split=”train”) dataset = FGVCAircraft(config_name=”manufacturer”, split=”train”)

BUILDER_CONFIGS: list = [BuilderConfig(name='variant', version=None, description='100 aircraft model variants (finest granularity)'), BuilderConfig(name='family', version=None, description='70 aircraft families (medium granularity)'), BuilderConfig(name='manufacturer', version=None, description='30 aircraft manufacturers (coarsest granularity)')]

DEFAULT_CONFIG_NAME: str | None = 'variant'

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/', assets=mappingproxy({'train': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/archives/fgvc-aircraft-2013b.tar.gz', fallbacks=[], checksum=None, filename=None), 'validation': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/archives/fgvc-aircraft-2013b.tar.gz', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/archives/fgvc-aircraft-2013b.tar.gz', fallbacks=[], checksum=None, filename=None)}), citation='@techreport{maji13fine-grained,\n                        title         = {Fine-Grained Visual Classification of Aircraft},\n                        author        = {S. Maji and J. Kannala and E. Rahtu and M. Blaschko and A. Vedaldi},\n                        year          = {2013},\n                        archivePrefix = {arXiv},\n                        eprint        = {1306.5151},\n                        primaryClass  = "cs.CV",\n                    }', license='Unknown', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.flowers102 module

class Flowers102(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Flowers102 Dataset

Abstract The Flowers102 dataset is a fine-grained image classification benchmark consisting of 102 flower categories commonly found in the United Kingdom. It was created to address the challenge of classifying objects with large intra-class variability and small inter-class differences. Each category contains between 40 and 258 images, totaling 8,189 images.

Context Fine-grained visual categorization (FGVC) focuses on differentiating between similar sub-categories of objects (e.g., different species of flowers or birds). Flowers102 serves as a standard benchmark in this domain. Unlike general object recognition (e.g., CIFAR-10), where classes are visually distinct (car vs. dog), Flowers102 requires models to learn subtle features like petal shape, texture, and color patterns.

Content The dataset consists of: - Images: 8,189 images stored in a single archive. - Labels: A MATLAB file mapping each image to one of 102 classes (0-101). - Splits: A predefined split ID file dividing the data into Training (1,020 images), Validation (1,020 images), and Test (6,149 images).

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.robots.ox.ac.uk/~vgg/data/flowers/102/', assets=mappingproxy({'images': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/flowers/102/102flowers.tgz', fallbacks=[], checksum=None, filename=None), 'labels': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/flowers/102/imagelabels.mat', fallbacks=[], checksum=None, filename=None), 'setid': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/flowers/102/setid.mat', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{nilsback2008flowers102,\n                         title={Automated flower classification over a large number of classes},\n                         author={Nilsback, Maria-Elena and Zisserman, Andrew},\n                         booktitle={2008 Sixth Indian conference on computer vision, graphics \\& image processing},\n                         pages={722--729},\n                         year={2008},\n                         organization={IEEE}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.flowers17 module

class Flowers17(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Oxford Flowers-17 classification dataset.

This set contains images of flowers belonging to 17 different categories. The images were acquired by searching the web and taking pictures. There are 80 images for each category.

There are three official partitions, each contains 680 training, 340 validation, and 340 test images. Select a partition with config_name=”partition1”, “partition2”, or “partition3”.

BUILDER_CONFIGS: list = [BuilderConfig(name='partition1', version=None, description='Official partition 1'), BuilderConfig(name='partition2', version=None, description='Official partition 2'), BuilderConfig(name='partition3', version=None, description='Official partition 3')]

DEFAULT_CONFIG_NAME: str | None = 'partition1'

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.robots.ox.ac.uk/~vgg/data/flowers/17/', assets=mappingproxy({'images': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/flowers/17/17flowers.tgz', fallbacks=[], checksum=None, filename=None), 'splits': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/flowers/17/datasplits.mat', fallbacks=[], checksum=None, filename=None)}), citation='@InProceedings{Nilsback06,\n            author = {Maria-Elena Nilsback and Andrew Zisserman},\n            title = {A Visual Vocabulary for Flower Classification},\n            booktitle = {IEEE Conference on Computer Vision and Pattern Recognition},\n            volume = {2},\n            pages = {1447--1454},\n            year = {2006},\n        }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.fmd module

Flickr Material Database with the released region masks preserved losslessly.

class FMD(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Load all 1,000 image/mask pairs in the unsplit train container.

The released JPEG masks decode as L or RGB and need not be binary. Their original decoded mode, size, and pixels are preserved as PNG, without grayscale conversion, thresholding, or resizing. Classification uses image and label only; experimental train/validation/test splits belong downstream.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://people.csail.mit.edu/lavanya/fmd.html', assets=mappingproxy({'train': DownloadInfo(url='https://people.csail.mit.edu/celiu/CVPR2010/FMD/FMD.zip', fallbacks=[], checksum='sha256:254475ddb37bf0fe1383ff9ea61eb71042db33af1f9a9aa8747391a72135b8f8', filename=None)}), citation='@article{sharan2014accuracy,\n  title={Accuracy and speed of material categorization in real-world images},\n  author={Sharan, Lavanya and Rosenholtz, Ruth and Adelson, Edward H.},\n  journal={Journal of Vision},\n  volume={14},\n  number={9},\n  pages={12},\n  year={2014},\n  doi={10.1167/14.9.12}\n}', license='See the source page for Creative Commons terms, image credits, and differently licensed exceptions.', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.food101 module

class Food101(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Food-101 Dataset.

The Food-101 dataset consists of 101 food categories with 101,000 images. For each class, 250 manually reviewed test images are provided as well as 750 training images.

Split sizes: - train: 75,750 images (750 images × 101 classes) - test: 25,250 images (250 images × 101 classes)

All images are automatically rescaled to have a maximum side length of 512 pixels.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://data.vision.ee.ethz.ch/cvl/datasets_extra/food-101/', assets=mappingproxy({'train': DownloadInfo(url='https://huggingface.co/datasets/haodoz0118/food101-img/resolve/main/food101_train.tar', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://huggingface.co/datasets/haodoz0118/food101-img/resolve/main/food101_test.tar', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{bossard14,\n            title = {Food-101 -- Mining Discriminative Components with Random Forests},\n            author = {Bossard, Lukas and Guillaumin, Matthieu and Van Gool, Luc},\n            booktitle = {European Conference on Computer Vision},\n            year = {2014}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.freiburg_groceries module

Freiburg Groceries with the authors’ five published train/test partitions.

class FreiburgGroceries(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Load author partitions split0 (default) through split4.

Each training image is returned once, in order of its first appearance in the author’s manifest. Train/test membership and labels follow the author partitions, without class balancing. There is no official validation split.

BUILDER_CONFIGS: list = [BuilderConfig(name='split0', version=None, description=''), BuilderConfig(name='split1', version=None, description=''), BuilderConfig(name='split2', version=None, description=''), BuilderConfig(name='split3', version=None, description=''), BuilderConfig(name='split4', version=None, description='')]

DEFAULT_CONFIG_NAME: str | None = 'split0'

LABELS = ('CANDY', 'JUICE', 'VINEGAR', 'OIL', 'CHOCOLATE', 'PASTA', 'RICE', 'MILK', 'SPICES', 'HONEY', 'JAM', 'NUTS', 'CHIPS', 'SODA', 'COFFEE', 'BEANS', 'TEA', 'CORN', 'CEREAL', 'CAKE', 'SUGAR', 'WATER', 'FLOUR', 'TOMATO_SAUCE', 'FISH')

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='http://www2.informatik.uni-freiburg.de/~eitel/freiburg_groceries_dataset.html', assets=mappingproxy({'images': DownloadInfo(url='http://aisdatasets.informatik.uni-freiburg.de/freiburg_groceries_dataset/freiburg_groceries_dataset.tar.gz', fallbacks=[], checksum='sha256:35c18d03bb6ed057f12a3c5aea0ac5eef9b8e1d5965d1e32142b7d1f1469e3f0', filename=None), 'annotations': DownloadInfo(url='https://codeload.github.com/PhilJd/freiburg_groceries_dataset/tar.gz/8827d2f61c2838756a069562b1c2ff53568ae33b', fallbacks=[], checksum='sha256:b07ba6e6c3760cb909f4b92de41f7ab505c3a910f8736613ea4b9712537aafb9', filename='freiburg-groceries-splits.tar.gz')}), citation='@article{jund2016groceries,\n    title={The Freiburg Groceries Dataset},\n    author={Jund, Philipp and Abdo, Nichola and Eitel, Andreas and Burgard, Wolfram},\n    journal={arXiv preprint arXiv:1611.05799},\n    year={2016}\n}', license="Research purposes only; commercial use requires the authors' consent. Cite Jund et al. (2016).", checksums=None)

SPLITS_REVISION = '8827d2f61c2838756a069562b1c2ff53568ae33b'

VERSION: Version = Version('1.0.0')

stable_datasets.images.fruits360 module

class Fruits360(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Fruits-360: 100x100 images of fruits, vegetables, nuts, seeds and mushrooms.

Each object was filmed while rotating on a slow motor in front of a white background, and the background was removed with a flood-fill algorithm. Different varieties of the same fruit are separate classes. The official Training/Test partitions are kept as released.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/fruits-360', assets=mappingproxy({'train': DownloadInfo(url='https://github.com/fruits-360/fruits-360-100x100/archive/18fbaa1b48a70d356e6aa47a8ff88156fb96b887.zip', fallbacks=[], checksum='sha256:50213ec3770a9f5dbceac91e8ca660b5507accb405ba89ac5fe9253b5597cadb', filename=None), 'test': DownloadInfo(url='https://github.com/fruits-360/fruits-360-100x100/archive/18fbaa1b48a70d356e6aa47a8ff88156fb96b887.zip', fallbacks=[], checksum='sha256:50213ec3770a9f5dbceac91e8ca660b5507accb405ba89ac5fe9253b5597cadb', filename=None)}), citation='@article{muresan2018fruit,\n                title   = {Fruit recognition from images using deep learning},\n                author  = {Mure{\\c{s}}an, Horea and Oltean, Mihai},\n                journal = {Acta Universitatis Sapientiae, Informatica},\n                volume  = {10},\n                number  = {1},\n                pages   = {26--42},\n                year    = {2018} }\n        ', license='', checksums=None)

VERSION: Version = Version('2026.10.18')

stable_datasets.images.galaxy10 module

class Galaxy10Decal(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Galaxy10 DECaLS Dataset: Galaxy morphology classification with DECaLS images.

Galaxy10 DECaLS is a much improved version of the original Galaxy10 dataset. It contains 17,736 256x256 pixel colored galaxy images (g, r and z band) separated into 10 classes. The images come from DESI Legacy Imaging Surveys (DECaLS) and labels come from Galaxy Zoo.

The original Galaxy10 dataset was created with Galaxy Zoo (GZ) Data Release 2 where volunteers classify ~270k of SDSS galaxy images. GZ later utilized images from DESI Legacy Imaging Surveys (DECaLS) with much better resolution and image quality. Galaxy10 DECaLS has combined all three (GZ DR2 with DECaLS images instead of SDSS images and DECaLS campaign a/b, c) resulting in ~441k of unique galaxies covered by DECaLS where ~18k of those images were selected in 10 broad classes using volunteer votes with more rigorous filtering.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://astronn.readthedocs.io/en/latest/galaxy10.html', assets=mappingproxy({'train': DownloadInfo(url='https://zenodo.org/records/10845026/files/Galaxy10_DECals.h5', fallbacks=[], checksum=None, filename=None)}), citation='@article{walmsley2020galaxy,\n                        title={Galaxy Zoo: probabilistic morphology through Bayesian CNNs and active learning},\n                        author={Walmsley, Mike and Smith, Lewis and Lintott, Chris and Gal, Yarin and Bamford, Steven and Dickinson, Hugh and Fortson, Lucy and Kruk, Sandor and Masters, Karen and Scarlata, Claudia and others},\n                        journal={Monthly Notices of the Royal Astronomical Society},\n                        volume={491},\n                        number={2},\n                        pages={1554--1574},\n                        year={2020},\n                        publisher={Oxford University Press}\n                    }', license='MIT', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.grocery_store module

class GroceryStore(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Grocery Store dataset: natural images of fruits, vegetables and packaged goods.

Smartphone photos taken in grocery stores, labeled with 81 fine-grained classes (e.g. “Granny-Smith”) that group into 43 coarse-grained classes (e.g. “Apple”). The official split has 2,640 train, 296 validation and 2,485 test images. Note that the validation set only covers 60 of the 81 fine-grained classes.

Each example has the image, the fine-grained label and the coarse_label.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/marcusklasson/GroceryStoreDataset', assets=mappingproxy({'train': DownloadInfo(url='https://github.com/marcusklasson/GroceryStoreDataset/archive/fc80ba90f803d79d0383df52c5a4ac5de99ff6fc.zip', fallbacks=[], checksum=None, filename='GroceryStoreDataset.zip'), 'validation': DownloadInfo(url='https://github.com/marcusklasson/GroceryStoreDataset/archive/fc80ba90f803d79d0383df52c5a4ac5de99ff6fc.zip', fallbacks=[], checksum=None, filename='GroceryStoreDataset.zip'), 'test': DownloadInfo(url='https://github.com/marcusklasson/GroceryStoreDataset/archive/fc80ba90f803d79d0383df52c5a4ac5de99ff6fc.zip', fallbacks=[], checksum=None, filename='GroceryStoreDataset.zip')}), citation='@inproceedings{klasson2019hierarchical,\n                         title={A Hierarchical Grocery Store Image Dataset with Visual and Semantic Labels},\n                         author={Klasson, Marcus and Zhang, Cheng and Kjellstr{\\"o}m, Hedvig},\n                         booktitle={IEEE Winter Conference on Applications of Computer Vision (WACV)},\n                         year={2019}}', license='MIT License', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.gtsrb module

Official German Traffic Sign Recognition Benchmark (GTSRB).

class GTSRB(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

GTSRB classification with the official final train and labeled test splits.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/houbensebastian/GermanTrafficSignBenchmarks', assets=mappingproxy({'train': DownloadInfo(url='https://sid.erda.dk/public/archives/daaeac0d7ce1152aea9b61d9f1e19370/GTSRB_Final_Training_Images.zip', fallbacks=['https://web.archive.org/web/20220418123022id_/https://sid.erda.dk/public/archives/daaeac0d7ce1152aea9b61d9f1e19370/GTSRB_Final_Training_Images.zip'], checksum='sha256:d32ac4b5fa9a1cbd1994768413902e8193599d9434cf0a8eb9cfd00a6d3a290c', filename=None), 'test_images': DownloadInfo(url='https://sid.erda.dk/public/archives/daaeac0d7ce1152aea9b61d9f1e19370/GTSRB_Final_Test_Images.zip', fallbacks=['https://web.archive.org/web/20220418123019id_/https://sid.erda.dk/public/archives/daaeac0d7ce1152aea9b61d9f1e19370/GTSRB_Final_Test_Images.zip'], checksum='sha256:48ba6fab7e877eb64eaf8de99035b0aaecfbc279bee23e35deca4ac1d0a837fa', filename=None), 'test_ground_truth': DownloadInfo(url='https://sid.erda.dk/public/archives/daaeac0d7ce1152aea9b61d9f1e19370/GTSRB_Final_Test_GT.zip', fallbacks=['https://web.archive.org/web/20220418123054id_/https://sid.erda.dk/public/archives/daaeac0d7ce1152aea9b61d9f1e19370/GTSRB_Final_Test_GT.zip'], checksum='sha256:f94e5a7614d75845c74c04ddb26b8796b9e483f43541dd95dd5b726504e16d6d', filename=None)}), citation='@inproceedings{Stallkamp-IJCNN-2011,\n    author = {Johannes Stallkamp and Marc Schlipsing and Jan Salmen and Christian Igel},\n    title = {The {G}erman {T}raffic {S}ign {R}ecognition {B}enchmark: A multi-class classification competition},\n    booktitle = {IEEE International Joint Conference on Neural Networks},\n    pages = {1453--1460},\n    year = {2011}\n}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.hasy_v2 module

class HASYv2(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

HASYv2 Dataset

Abstract The HASYv2 dataset contains handwritten symbol images of 369 classes. It includes over 168,000 samples categorized into various classes like Latin characters, numerals, and symbols. Each image is 32x32 pixels in size. The dataset was created to benchmark the classification of mathematical symbols and handwritten characters.

Context Recognizing handwritten mathematical symbols is a challenging task due to the similarity between classes (e.g., ‘1’, ‘l’, ‘|’) and the large number of unique symbols used in scientific notation. HASYv2 serves as a standard benchmark for testing classifiers on a large number of classes (369) with low resolution (32x32).

Content The dataset consists of: - Images: 168,236 black-and-white images (32x32 pixels). - Labels: 369 distinct classes. - Splits: The dataset includes 10 pre-defined folds. This implementation uses ‘Fold 1’ as the standard train/test split.

BUILDER_CONFIGS: list = [BuilderConfig(name='fold-1', version=Version('1.0.0'), description='HASYv2 dataset using fold 1 as the test set.'), BuilderConfig(name='fold-2', version=Version('1.0.0'), description='HASYv2 dataset using fold 2 as the test set.'), BuilderConfig(name='fold-3', version=Version('1.0.0'), description='HASYv2 dataset using fold 3 as the test set.'), BuilderConfig(name='fold-4', version=Version('1.0.0'), description='HASYv2 dataset using fold 4 as the test set.'), BuilderConfig(name='fold-5', version=Version('1.0.0'), description='HASYv2 dataset using fold 5 as the test set.'), BuilderConfig(name='fold-6', version=Version('1.0.0'), description='HASYv2 dataset using fold 6 as the test set.'), BuilderConfig(name='fold-7', version=Version('1.0.0'), description='HASYv2 dataset using fold 7 as the test set.'), BuilderConfig(name='fold-8', version=Version('1.0.0'), description='HASYv2 dataset using fold 8 as the test set.'), BuilderConfig(name='fold-9', version=Version('1.0.0'), description='HASYv2 dataset using fold 9 as the test set.'), BuilderConfig(name='fold-10', version=Version('1.0.0'), description='HASYv2 dataset using fold 10 as the test set.')]

DEFAULT_CONFIG_NAME: str | None = 'fold-1'

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/MartinThoma/HASY', assets=mappingproxy({'train': DownloadInfo(url='https://zenodo.org/record/259444/files/HASYv2.tar.bz2?download=1', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://zenodo.org/record/259444/files/HASYv2.tar.bz2?download=1', fallbacks=[], checksum=None, filename=None)}), citation='@article{thoma2017hasyv2,\n                         title={The hasyv2 dataset},\n                         author={Thoma, Martin},\n                         journal={arXiv preprint arXiv:1701.08380},\n                         year={2017}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.horse_or_human module

class HorseOrHuman(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Horses vs. Humans binary image classification dataset.

Contains 300x300 RGB images belonging to two classes: horses and humans.

The dataset preserves its original predefined splits: - train: 1,027 images - test: 256 images from the original validation archive

Each example contains a PIL RGB image and an integer class label (0 for horses, 1 for humans).

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.tensorflow.org/datasets/catalog/horses_or_humans', assets=mappingproxy({'train': DownloadInfo(url='https://storage.googleapis.com/learning-datasets/horse-or-human.zip', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://storage.googleapis.com/learning-datasets/validation-horse-or-human.zip', fallbacks=[], checksum=None, filename=None)}), citation='@misc{horses_or_humans,\n        author = {Laurence Moroney},\n        title = {Horses or Humans Dataset},\n        month = feb,\n        year = {2019},\n        url = {http://laurencemoroney.com/horses-or-humans-dataset}\n        }', license='CC BY 2.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.imagenet_10 module

class Imagenette(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Imagenette: 10 easily classified classes from ImageNet.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/fastai/imagenette', assets=mappingproxy({'archive': DownloadInfo(url='https://s3.amazonaws.com/fast-ai-imageclas/imagenette2.tgz', fallbacks=[], checksum=None, filename=None)}), citation='@misc{howard2019imagenette,\n            author={Jeremy Howard},\n            title={Imagenette: A smaller subset of 10 easily classified classes from ImageNet},\n            year={2019},\n            url={https://github.com/fastai/imagenette}\n        }', license='', checksums=None)

VERSION: Version = Version('3.0.0')

stable_datasets.images.imagenet_100 module

class ImageNet100(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: ImageNet1K

ImageNet-100: the first 100 wnids of canonical sorted-alphabetical ImageNet-1K.

Labels are integers in [0, 100), where label i corresponds to IN100_CLASSES[i].

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.image-net.org/challenges/LSVRC/2012/', assets=mappingproxy({'train': DownloadInfo(url='https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_train.tar', fallbacks=[], checksum=None, filename=None), 'val': DownloadInfo(url='https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_val.tar', fallbacks=[], checksum=None, filename=None), 'devkit': DownloadInfo(url='https://image-net.org/data/ILSVRC/2012/ILSVRC2012_devkit_t12.tar.gz', fallbacks=[], checksum=None, filename=None)}), citation='@article{deng2009imagenet,\n        title={ImageNet: A large-scale hierarchical image database},\n        author={Deng, Jia and others},\n        journal={CVPR},\n        year={2009}\n    }', license='', checksums=None)

VERSION: Version = Version('3.0.0')

stable_datasets.images.imagenet_1k module

class ImageNet1K(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: _ImageNetArchiveMixin, BaseDatasetBuilder

ImageNet-1K (ILSVRC 2012): 1,000 classes, built from the official train/val tars and devkit.

Labels are integers in [0, 1000), where label i corresponds to IN1K_CLASSES[i] (wnids sorted alphabetically).

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.image-net.org/challenges/LSVRC/2012/', assets=mappingproxy({'train': DownloadInfo(url='https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_train.tar', fallbacks=[], checksum=None, filename=None), 'val': DownloadInfo(url='https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_val.tar', fallbacks=[], checksum=None, filename=None), 'devkit': DownloadInfo(url='https://image-net.org/data/ILSVRC/2012/ILSVRC2012_devkit_t12.tar.gz', fallbacks=[], checksum=None, filename=None)}), citation='@article{deng2009imagenet,\n        title={ImageNet: A large-scale hierarchical image database},\n        author={Deng, Jia and others},\n        journal={CVPR},\n        year={2009}\n    }', license='', checksums=None)

VERSION: Version = Version('3.0.0')

stable_datasets.images.imagenet_a module

ImageNet-A: natural adversarial examples for 200 ImageNet classes.

class ImageNetA(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

ImageNet-A: 7,500 real-world photos that standard ImageNet classifiers get wrong.

ImageNet-A is an evaluation-only robustness benchmark, so it has a single test split. Labels are integers in [0, 200): label i is the class IMAGENET_A_CLASS_NAMES[i] with WordNet ID IMAGENET_A_WNIDS[i], and IMAGENET_A_TO_IN1K[i] is its ImageNet-1K index.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/hendrycks/natural-adv-examples', assets=mappingproxy({'test': DownloadInfo(url='https://people.eecs.berkeley.edu/~hendrycks/imagenet-a.tar', fallbacks=[], checksum='sha256:3bb3632277e6ba6392ea64c02ddbf4dd2266c9caffd6bc09c9656d28f012589e', filename=None)}), citation='@article{hendrycks2021nae,\n                      title={Natural Adversarial Examples},\n                      author={Dan Hendrycks and Kevin Zhao and Steven Basart and Jacob Steinhardt and Dawn Song},\n                      journal={CVPR},\n                      year={2021}}', license='MIT', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.imagenet_r module

ImageNet-R: renditions (art, cartoons, sketches, toys, …) of 200 ImageNet classes.

class ImageNetR(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

ImageNet-R: 30,000 renditions (art, cartoons, sketches, toys, …) of 200 ImageNet classes.

ImageNet-R is an evaluation-only robustness benchmark, so it has a single test split. Labels are integers in [0, 200): label i is the class IMAGENET_R_CLASS_NAMES[i] with WordNet ID IMAGENET_R_WNIDS[i], and IMAGENET_R_TO_IN1K[i] is its ImageNet-1K index.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/hendrycks/imagenet-r', assets=mappingproxy({'test': DownloadInfo(url='https://people.eecs.berkeley.edu/~hendrycks/imagenet-r.tar', fallbacks=[], checksum='sha256:18c6bf493b39a0d975d48e587437f562caab9c52ae6327dcfa9dd8eb54aa1b52', filename=None)}), citation='@article{hendrycks2021many,\n                      title={The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization},\n                      author={Dan Hendrycks and Steven Basart and Norman Mu and Saurav Kadavath and Frank Wang and Evan Dorundo and Rahul Desai and Tyler Zhu and Samyak Parajuli and Mike Guo and Dawn Song and Jacob Steinhardt and Justin Gilmer},\n                      journal={ICCV},\n                      year={2021}}', license='MIT', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.imagenet_sketch module

ImageNet-Sketch: black-and-white sketches of the 1000 ImageNet-1K classes.

class ImageNetSketch(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Test-only sketch images, labeled in the ImageNet-1K class order.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/HaohanWang/ImageNet-Sketch', assets=mappingproxy({'test': DownloadInfo(url='https://huggingface.co/datasets/songweig/imagenet_sketch/resolve/main/data/ImageNet-Sketch.zip', fallbacks=[], checksum='sha256:4d3ce6b82a7c48aabf14aeac105eb533e75707f152cc858f55bfc022031a72ed', filename=None)}), citation='@inproceedings{wang2019learning,\n                      title={Learning Robust Global Representations by Penalizing Local Predictive Power},\n                      author={Wang, Haohan and Ge, Songwei and Lipton, Zachary and Xing, Eric P},\n                      booktitle={Advances in Neural Information Processing Systems},\n                      pages={10506--10518},\n                      year={2019}}', license='MIT', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.imagenet_v2 module

ImageNetV2: three new ImageNet-1K test sets (Recht et al., ICML 2019).

class ImageNetV2(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Test-only ImageNet-1K test sets collected anew, labeled in the ImageNet-1K class order.

BUILDER_CONFIGS: list = [ImageNetV2Config(name='matched-frequency', version=Version('1.0.0'), description='ImageNetV2 MatchedFrequency (imagenetv2-b-33)'), ImageNetV2Config(name='threshold-0.7', version=Version('1.0.0'), description='ImageNetV2 Threshold0.7 (imagenetv2-a-44)'), ImageNetV2Config(name='top-images', version=Version('1.0.0'), description='ImageNetV2 TopImages (imagenetv2-c-12)')]

DEFAULT_CONFIG_NAME: str | None = 'matched-frequency'

VERSION: Version = Version('1.0.0')

class ImageNetV2Config(*, archive: str, checksum: str, **kwargs)

Bases: BuilderConfig

BuilderConfig naming the archive of one ImageNetV2 variant and its SHA-256.

stable_datasets.images.imagenette module

class Imagenette(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Imagenette (ImageNet-10) from FastAI’s public tarball.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/fastai/imagenette', assets=mappingproxy({'archive': DownloadInfo(url='https://s3.amazonaws.com/fast-ai-imageclas/imagenette2.tgz', fallbacks=[], checksum=None, filename=None)}), citation='@misc{howard2019imagenette,\n            author={Jeremy Howard},\n            title={Imagenette: A smaller subset of 10 easily classified classes from ImageNet},\n            year={2019},\n            url={https://github.com/fastai/imagenette}\n        }', license='', checksums=None)

VERSION: Version = Version('2.0.0')

stable_datasets.images.imagewoof module

Full-resolution Imagewoof with original labels and official splits.

class Imagewoof(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Ten dog classes from ImageNet, using the official Imagewoof v2 archive.

BUILDER_CONFIGS: list = [BuilderConfig(name='default', version=None, description='Full-resolution Imagewoof v2, original labels.')]

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/fastai/imagenette#imagewoof', assets=mappingproxy({'train': DownloadInfo(url='https://s3.amazonaws.com/fast-ai-imageclas/imagewoof2.tgz', fallbacks=[], checksum=None, filename=None), 'val': DownloadInfo(url='https://s3.amazonaws.com/fast-ai-imageclas/imagewoof2.tgz', fallbacks=[], checksum=None, filename=None)}), citation="@software{Howard_Imagewoof_2019,\n            title={Imagewoof: a subset of 10 classes from Imagenet that aren't so easy to classify},\n            author={Jeremy Howard},\n            year={2019},\n            month={March},\n            publisher={GitHub},\n            url={https://github.com/fastai/imagenette#imagewoof}\n        }", license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.intel_image_classification module

class IntelImageClassification(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Load Intel scene images with a held-out test set and stratified validation split.

CLASS_NAMES = ('buildings', 'forest', 'glacier', 'mountain', 'sea', 'street')

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.kaggle.com/datasets/puneet6060/intel-image-classification', assets=mappingproxy({'train': DownloadInfo(url='https://www.kaggle.com/api/v1/datasets/download/puneet6060/intel-image-classification', fallbacks=[], checksum=None, filename='intel-image-classification.zip'), 'validation': DownloadInfo(url='https://www.kaggle.com/api/v1/datasets/download/puneet6060/intel-image-classification', fallbacks=[], checksum=None, filename='intel-image-classification.zip'), 'test': DownloadInfo(url='https://www.kaggle.com/api/v1/datasets/download/puneet6060/intel-image-classification', fallbacks=[], checksum=None, filename='intel-image-classification.zip')}), citation='@misc{bansal2019intel,\n        author={Bansal, Puneet},\n        title={Intel Image Classification},\n        year={2019},\n        howpublished={Kaggle},\n        url={https://www.kaggle.com/datasets/puneet6060/intel-image-classification}\n        }', license='', checksums=None)

SPLIT_SEED = 42

VALIDATION_FRACTION = 0.1

VERSION: Version = Version('1.0.0')

stable_datasets.images.isic2016 module

ISIC 2016 Task 3 original image classification data and official labels.

class ISIC2016(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

ISIC 2016 Task 3 dermoscopic skin lesion classification dataset.

Contains 900 training and 379 test JPEG images with variable dimensions. Labels identify benign lesions (0) and malignant lesions (1). Images and labels were released for the ISBI 2016 skin lesion analysis challenge; the official train/test partitions are retained.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://challenge.isic-archive.com/landing/2016/39/', assets=mappingproxy({'Training_Data.zip': DownloadInfo(url='https://isic-archive.s3.amazonaws.com/challenges/2016/ISBI2016_ISIC_Part3_Training_Data.zip', fallbacks=[], checksum='sha256:d78bd1de08511b514e58784c91fd2d8dab24e407cafd023a7debbceb9155ffb2', filename=None), 'Training_GroundTruth.csv': DownloadInfo(url='https://isic-archive.s3.amazonaws.com/challenges/2016/ISBI2016_ISIC_Part3_Training_GroundTruth.csv', fallbacks=[], checksum='sha256:0ffadd63b7cec650cddfbe5df54d1d3167453fe6004c2bbf05824184b0c4d405', filename=None), 'Test_Data.zip': DownloadInfo(url='https://isic-archive.s3.amazonaws.com/challenges/2016/ISBI2016_ISIC_Part3_Test_Data.zip', fallbacks=[], checksum='sha256:7c52d5c013b6a79b92c246d48d9f175299670537c128bcc1400b35891f6350e6', filename=None), 'Test_GroundTruth.csv': DownloadInfo(url='https://isic-archive.s3.amazonaws.com/challenges/2016/ISBI2016_ISIC_Part3_Test_GroundTruth.csv', fallbacks=[], checksum='sha256:8f7718bfe5ed110e0235b0d9afc40fe89e52610c5687cc9cd7dd0c23352b82ce', filename=None)}), citation='@article{gutman2016skin,\n            title={Skin Lesion Analysis toward Melanoma Detection: A Challenge at the International\n                   Symposium on Biomedical Imaging (ISBI) 2016, hosted by the International Skin\n                   Imaging Collaboration (ISIC)},\n            author={Gutman, David and Codella, Noel C. F. and Celebi, Emre and Helba, Brian\n                    and Marchetti, Michael and Mishra, Nabin and Halpern, Allan},\n            journal={arXiv preprint arXiv:1605.01397}, year={2016}}\n        ', license='CC0-1.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.k_mnist module

class KMNIST(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Image classification. The Kuzushiji-MNIST dataset consists of 70,000 28x28 grayscale images of 10 classes of Kuzushiji (cursive Japanese) characters, with 7,000 images per class. There are 60,000 training images and 10,000 test images. Kuzushiji-MNIST is a drop-in replacement for the MNIST dataset, providing a more challenging alternative for benchmarking machine learning algorithms.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='http://codh.rois.ac.jp/kmnist/', assets=mappingproxy({'train': DownloadInfo(url='https://codh.rois.ac.jp/kmnist/dataset/kmnist/kmnist-train-imgs.npz', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://codh.rois.ac.jp/kmnist/dataset/kmnist/kmnist-test-imgs.npz', fallbacks=[], checksum=None, filename=None)}), citation='@online{clanuwat2018deep,\n                         author       = {Tarin Clanuwat and Mikel Bober-Irizar and Asanobu Kitamoto and Alex Lamb and Kazuaki Yamamoto and David Ha},\n                         title        = {Deep Learning for Classical Japanese Literature},\n                         date         = {2018-12-03},\n                         year         = {2018},\n                         eprintclass  = {cs.CV},\n                         eprinttype   = {arXiv},\n                         eprint       = {cs.CV/1812.01718}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.k_mnist49 module

class KMNIST49(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Image classification. The Kuzushiji-49 dataset consists of 270,912 28x28 grayscale images of 49 classes of Kuzushiji (cursive Japanese) hiragana characters. There are 232,365 training images and 38,547 test images. Unlike Kuzushiji-MNIST, the classes are highly imbalanced, ranging from a few hundred to several thousand images per class.

LABEL_URLS = {'test': 'https://codh.rois.ac.jp/kmnist/dataset/k49/k49-test-labels.npz', 'train': 'https://codh.rois.ac.jp/kmnist/dataset/k49/k49-train-labels.npz'}

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='http://codh.rois.ac.jp/kmnist/', assets=mappingproxy({'train': DownloadInfo(url='https://codh.rois.ac.jp/kmnist/dataset/k49/k49-train-imgs.npz', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://codh.rois.ac.jp/kmnist/dataset/k49/k49-test-imgs.npz', fallbacks=[], checksum=None, filename=None)}), citation='@online{clanuwat2018deep,\n                         author       = {Tarin Clanuwat and Mikel Bober-Irizar and Asanobu Kitamoto and Alex Lamb and Kazuaki Yamamoto and David Ha},\n                         title        = {Deep Learning for Classical Japanese Literature},\n                         date         = {2018-12-03},\n                         year         = {2018},\n                         eprintclass  = {cs.CV},\n                         eprinttype   = {arXiv},\n                         eprint       = {cs.CV/1812.01718}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.kannada_mnist module

class KannadaMNIST(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Kannada handwritten digits, using the original 60,000/10,000 split.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/vinayprabhu/Kannada_MNIST', assets=mappingproxy({'train_images': DownloadInfo(url='https://raw.githubusercontent.com/vinayprabhu/Kannada_MNIST/master/data/output_tensors/MNIST_format/X_kannada_MNIST_train-idx3-ubyte.gz', fallbacks=[], checksum='sha256:11785e1e0846a852266d5323a51c1b495c207ad2acbeaaaf108211163481b212', filename=None), 'train_labels': DownloadInfo(url='https://raw.githubusercontent.com/vinayprabhu/Kannada_MNIST/master/data/output_tensors/MNIST_format/y_kannada_MNIST_train-idx1-ubyte.gz', fallbacks=[], checksum='sha256:ef11e1d3e378fefbe78d4c9913dc9617c83fc5f9240b090121fcc335b5642201', filename=None), 'test_images': DownloadInfo(url='https://raw.githubusercontent.com/vinayprabhu/Kannada_MNIST/master/data/output_tensors/MNIST_format/X_kannada_MNIST_test-idx3-ubyte.gz', fallbacks=[], checksum='sha256:4628f3fa7b1d32137449dd356be4601da0c28c420b79f1b745a5134504ca35ca', filename=None), 'test_labels': DownloadInfo(url='https://raw.githubusercontent.com/vinayprabhu/Kannada_MNIST/master/data/output_tensors/MNIST_format/y_kannada_MNIST_test-idx1-ubyte.gz', fallbacks=[], checksum='sha256:645db5902d290681c9e5b452428e24636963a50a3036c2201240351b88786272', filename=None)}), citation='@article{prabhu2019kannada,\n            title={Kannada-MNIST: A new handwritten digits dataset for the Kannada language},\n            author={Prabhu, Vinay Uday},\n            journal={arXiv preprint arXiv:1908.01242}, year={2019}}\n        ', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.kather_colorectal_histology module

The eight-class Kather colorectal histology texture dataset.

class KatherColorectalHistology(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Balanced classification of 5,000 colorectal histology tissue tiles from the Kather dataset. The dataset contains 625 examples in each of eight classes.

Only train is provided: the release has no official train/test split. TIFF tiles are stored as RGB PNGs without resizing. source_image_id preserves the source image name embedded in each tile’s filename, while slide_id is its numeric CRC-Prim-HE-NN prefix, grouping letter-suffixed variants (for example, CRC-Prim-HE-10c) under that prefix. These groups can be used to keep related tiles together when constructing splits; they are not independently verified patient identifiers.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://zenodo.org/records/53169', assets=mappingproxy({'train': DownloadInfo(url='https://zenodo.org/records/53169/files/Kather_texture_2016_image_tiles_5000.zip', fallbacks=[], checksum='md5:0ddbebfc56344752028fda72602aaade', filename=None)}), citation='@article{kather2016multi,\n            title={Multi-class texture analysis in colorectal cancer histology},\n            author={Kather, Jakob Nikolas and Weis, Cleo-Aron and Bianconi, Francesco\n                    and Melchers, Susanne M and Schad, Lothar R and Gaiser, Timo\n                    and Marx, Alexander and Z{\\"o}llner, Frank Gerrit},\n            journal={Scientific Reports},\n            volume={6},\n            pages={27988},\n            year={2016},\n            doi={10.1038/srep27988}\n        }', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.kitti_distance module

class KITTIDistance(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

KITTI Distance (closest-vehicle distance classification).

Abstract A 4-way image classification task built from the KITTI 2D/3D object detection benchmark: given a single front-camera frame from a car driving around Karlsruhe, predict how far away the closest vehicle (car, van or truck) is. Classes are <= 8m, 8-20m, > 20m and no vehicle. It is the kitti/closest_vehicle_distance task of the Visual Task Adaptation Benchmark (VTAB-1k) and the “KITTI Distance” dataset of CLIP’s linear-probe and zero-shot evaluation suites.

Context Unlike most classification benchmarks, which test object identity, this task probes spatial / depth understanding on real driving footage, so it is a useful check of whether representations encode geometry.

Content

• Images: the 7,481 labelled frames of KITTI’s object training set (left colour camera, ~1242x375 PNG). KITTI’s official testing set has no public labels and is not used.

• Labels: derived from the 3D object annotations (label_2): the minimum camera-frame z over all Car / Van / Truck objects, binned with VTAB’s thresholds.

• Splits: frames are split by source drive (video), so no drive appears in more than one split, reproducing TFDS kitti (v3.x) exactly: 10% of drives for test, 10% for validation (seed 123), the rest for train. This gives 6,347 / 423 / 711 frames for train / validation / test.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.cvlibs.net/datasets/kitti/eval_object.php?obj_benchmark=3d', assets=mappingproxy({'images': DownloadInfo(url='https://s3.eu-central-1.amazonaws.com/avg-kitti/data_object_image_2.zip', fallbacks=[], checksum='sha256:351c5a2aa0cd9238b50174a3a62b846bc5855da256b82a196431d60ff8d43617', filename=None), 'labels': DownloadInfo(url='https://s3.eu-central-1.amazonaws.com/avg-kitti/data_object_label_2.zip', fallbacks=[], checksum='sha256:4efc76220d867e1c31bb980bbf8cbc02599f02a9cb4350effa98dbb04aaed880', filename=None), 'devkit': DownloadInfo(url='https://s3.eu-central-1.amazonaws.com/avg-kitti/devkit_object.zip', fallbacks=[], checksum='sha256:ce0b76b69c0c5f89690a0d65b7302bbbdb962a0c7e8aba6efc7050d1b04b4cf1', filename=None)}), citation='@inproceedings{geiger2012kitti,\n                    title={Are we ready for Autonomous Driving? The KITTI Vision Benchmark Suite},\n                    author={Geiger, Andreas and Lenz, Philip and Urtasun, Raquel},\n                    booktitle={Conference on Computer Vision and Pattern Recognition (CVPR)},\n                    year={2012}}\n                    @article{zhai2019vtab,\n                    title={A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmark},\n                    author={Zhai, Xiaohua and Puigcerver, Joan and Kolesnikov, Alexander and Ruyssen, Pierre and\n                            Riquelme, Carlos and Lucic, Mario and Djolonga, Josip and Pinto, Andre Susano and\n                            Neumann, Maxim and Dosovitskiy, Alexey and others},\n                    journal={arXiv preprint arXiv:1910.04867},\n                    year={2019}}', license='CC BY-NC-SA 3.0', checksums=None)

VERSION: Version = Version('1.0.0')

distance_to_label(distance: float) → int

Bin a closest-vehicle distance (metres) into the 4 VTAB classes.

stable_datasets.images.kth_tips module

class KTHTIPS(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

KTH-TIPS material images under varying scale, pose, and lighting.

LABELS = ['aluminium_foil', 'brown_bread', 'corduroy', 'cotton', 'cracker', 'linen', 'orange_peel', 'sandpaper', 'sponge', 'styrofoam']

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.csc.kth.se/cvap/databases/kth-tips/', assets=mappingproxy({'train': DownloadInfo(url='https://www.csc.kth.se/cvap/databases/kth-tips/kth_tips_col_200x200.tar', fallbacks=[], checksum=None, filename=None)}), citation='@misc{fritz2004kthtips,\n            title={The KTH-TIPS database},\n            author={Fritz, Mario and Hayman, Eric and Caputo, Barbara and Eklundh, Jan-Olof},\n            year={2004},\n            url={https://www.csc.kth.se/cvap/databases/kth-tips/}\n        }', license='Unspecified: no dataset-specific license identified in the official source. See the documentation for source links and usage information.', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.lc25000 module

class LC25000(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Lung and Colon Cancer Histopathological Image Dataset (LC25000)

This dataset has 25,000 color histopathology images of 768x768 pixels in five balanced classes of 5,000: colon adenocarcinoma, benign colon, lung adenocarcinoma, benign lung and lung squamous cell carcinoma. The images were made by rotating and flipping 1,250 original tiles.

There is no official split, so 20% of the LC25000-clean tile groups of each class are held out for test with a fixed seed. No tile is in both splits.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/tampapath/lung_colon_image_set', assets=mappingproxy({'train': DownloadInfo(url='https://www.kaggle.com/api/v1/datasets/download/andrewmvd/lung-and-colon-cancer-histopathological-images?datasetVersionNumber=1', fallbacks=[], checksum=None, filename='lc25000.zip'), 'test': DownloadInfo(url='https://www.kaggle.com/api/v1/datasets/download/andrewmvd/lung-and-colon-cancer-histopathological-images?datasetVersionNumber=1', fallbacks=[], checksum=None, filename='lc25000.zip')}), citation='@article{borkowski2019lc25000,\n                      title={Lung and Colon Cancer Histopathological Image Dataset (LC25000)},\n                      author={Borkowski, Andrew A and Bui, Marilyn M and Thomas, L Brannon and Wilson, Catherine P and DeLand, Lauren A and Mastorides, Stephen M},\n                      journal={arXiv preprint arXiv:1912.12142},\n                      year={2019}}\n                    @inproceedings{batchkala2024lc25000clean,\n                      title={Evaluating Histopathology Foundation Models for Few-Shot Tissue Clustering: An Application to LC25000 Augmented Dataset Cleaning},\n                      author={Batchkala, George and Li, Bin and Rittscher, Jens},\n                      booktitle={Data Engineering in Medical Imaging (DEMI), MICCAI Workshop},\n                      year={2024}}', license='CC BY-SA 4.0 (images), CC BY 4.0 (LC25000-clean groups)', checksums=None)

VERSION: Version = Version('1.0.0')

split_groups(group_rows)

Map each group id to “train” or “test”, holding out 20% of the groups of every class. All copies of one original tile share a group, so no tile appears in both splits.

stable_datasets.images.linnaeus5 module

class Linnaeus5(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Linnaeus 5 Dataset

Abstract The Linnaeus 5 dataset contains 1,600 RGB images sized 256x256 pixels, categorized into 5 classes: berry, bird, dog, flower, and other (negative set). It was created to benchmark fine-grained classification and object recognition tasks.

Context While many datasets focus on broad object categories (like CIFAR-10), Linnaeus 5 offers a focused challenge on specific natural objects plus a “negative” class (‘other’). It serves as a good middle-ground benchmark between simple digit recognition (MNIST) and large-scale natural image classification (ImageNet).

Content The dataset consists of: - Images: 8,000 color images (256x256 pixels). - Classes: 5 categories (berry, bird, dog, flower, other). - Splits: Pre-split into Training (1,200 images per class) and Test (400 images per class).

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='http://chaladze.com/l5/', assets=mappingproxy({'train': DownloadInfo(url='http://chaladze.com/l5/img/Linnaeus%205%20256X256.rar', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='http://chaladze.com/l5/img/Linnaeus%205%20256X256.rar', fallbacks=[], checksum=None, filename=None)}), citation='@article{chaladze2017linnaeus,\n                      title={Linnaeus 5 dataset for machine learning},\n                      author={Chaladze, G and Kalatozishvili, L},\n                      journal={chaladze.com},\n                      year={2017}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.malaria module

class Malaria(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Balanced classification of parasitized and uninfected red blood cells.

The NIH Malaria Cell Images dataset contains 27,558 variable-resolution RGB cell crops and has no official train/test split. patient_id preserves the identifier from NIH’s patient-to-cell mappings; source_image_id identifies the microscopy image before individual cells were cropped. Group by patient when constructing splits to avoid sharing a patient’s cells across them.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://lhncbc.nlm.nih.gov/LHC-research/LHC-projects/image-processing/malaria-project.html', assets=mappingproxy({'images': DownloadInfo(url='https://data.lhncbc.nlm.nih.gov/public/Malaria/cell_images.zip', fallbacks=[], checksum='sha256:0a949556b2414159b5100192609805376654c4266d8d187be9b1922fad43c668', filename=None), 'patient_mapping_parasitized': DownloadInfo(url='https://data.lhncbc.nlm.nih.gov/public/Malaria/patientid_cellmapping_parasitized.csv', fallbacks=[], checksum='sha256:d0367e513397404e980baee2a641bce9ce329a22e62ea9007962dfca2f8418d3', filename=None), 'patient_mapping_uninfected': DownloadInfo(url='https://data.lhncbc.nlm.nih.gov/public/Malaria/patientid_cellmapping_uninfected.csv', fallbacks=[], checksum='sha256:a8577b21e7154724f4bbd18326218e2c63a99b22ab372b79a7530d36df6dab78', filename=None)}), citation='@article{rajaraman2018pre,\n            title={Pre-trained convolutional neural networks as feature extractors toward\n                   improved malaria parasite detection in thin blood smear images},\n            author={Rajaraman, Sivaramakrishnan and Antani, Sameer K and Poostchi, Mahdieh\n                    and Silamut, Kamolrat and Hossain, Md A and Maude, Richard J and Jaeger,\n                    Stefan and Thoma, George R},\n            journal={PeerJ},\n            volume={6},\n            pages={e4568},\n            year={2018},\n            doi={10.7717/peerj.4568}\n        }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.med_mnist module

class MedMNIST(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

MedMNIST, a large-scale MNIST-like collection of standardized biomedical images, including 12 datasets for 2D and 6 datasets for 3D.

BUILDER_CONFIGS: list = [MedMNISTConfig(name='pathmnist', version=Version('1.0.0'), description='MedMNIST PathMNIST (2D)'), MedMNISTConfig(name='chestmnist', version=Version('1.0.0'), description='MedMNIST ChestMNIST (2D, multi-label)'), MedMNISTConfig(name='dermamnist', version=Version('1.0.0'), description='MedMNIST DermaMNIST (2D)'), MedMNISTConfig(name='octmnist', version=Version('1.0.0'), description='MedMNIST OCTMNIST (2D)'), MedMNISTConfig(name='pneumoniamnist', version=Version('1.0.0'), description='MedMNIST PneumoniaMNIST (2D)'), MedMNISTConfig(name='retinamnist', version=Version('1.0.0'), description='MedMNIST RetinaMNIST (2D)'), MedMNISTConfig(name='breastmnist', version=Version('1.0.0'), description='MedMNIST BreastMNIST (2D)'), MedMNISTConfig(name='bloodmnist', version=Version('1.0.0'), description='MedMNIST BloodMNIST (2D)'), MedMNISTConfig(name='tissuemnist', version=Version('1.0.0'), description='MedMNIST TissueMNIST (2D)'), MedMNISTConfig(name='organamnist', version=Version('1.0.0'), description='MedMNIST OrganAMNIST (2D)'), MedMNISTConfig(name='organcmnist', version=Version('1.0.0'), description='MedMNIST OrganCMNIST (2D)'), MedMNISTConfig(name='organsmnist', version=Version('1.0.0'), description='MedMNIST OrganSMNIST (2D)'), MedMNISTConfig(name='organmnist3d', version=Version('1.0.0'), description='MedMNIST OrganMNIST3D (3D)'), MedMNISTConfig(name='nodulemnist3d', version=Version('1.0.0'), description='MedMNIST NoduleMNIST3D (3D)'), MedMNISTConfig(name='adrenalmnist3d', version=Version('1.0.0'), description='MedMNIST AdrenalMNIST3D (3D)'), MedMNISTConfig(name='fracturemnist3d', version=Version('1.0.0'), description='MedMNIST FractureMNIST3D (3D)'), MedMNISTConfig(name='vesselmnist3d', version=Version('1.0.0'), description='MedMNIST VesselMNIST3D (3D)'), MedMNISTConfig(name='synapsemnist3d', version=Version('1.0.0'), description='MedMNIST SynapseMNIST3D (3D)')]

VERSION: Version = Version('1.0.0')

class MedMNISTConfig(*, num_classes: int, is_3d: bool = False, multi_label: bool = False, **kwargs)

Bases: BuilderConfig

BuilderConfig with per-variant metadata used by MedMNIST._info().

stable_datasets.images.mnist module

class MNIST(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

MNIST Dataset using raw IDX files for digit classification.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='http://yann.lecun.com/exdb/mnist/', assets=mappingproxy({'train_images': DownloadInfo(url='https://storage.googleapis.com/cvdf-datasets/mnist/train-images-idx3-ubyte.gz', fallbacks=[], checksum=None, filename=None), 'train_labels': DownloadInfo(url='https://storage.googleapis.com/cvdf-datasets/mnist/train-labels-idx1-ubyte.gz', fallbacks=[], checksum=None, filename=None), 'test_images': DownloadInfo(url='https://storage.googleapis.com/cvdf-datasets/mnist/t10k-images-idx3-ubyte.gz', fallbacks=[], checksum=None, filename=None), 'test_labels': DownloadInfo(url='https://storage.googleapis.com/cvdf-datasets/mnist/t10k-labels-idx1-ubyte.gz', fallbacks=[], checksum=None, filename=None)}), citation='@misc{lecun1998mnist,\n                          author={Yann LeCun and Corinna Cortes and Christopher J.C. Burges},\n                          title={The MNIST database of handwritten digits},\n                          year={1998},\n                          url={http://yann.lecun.com/exdb/mnist/}\n                        }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.mnist_c module

class MNISTC(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

MNIST-C: a corrupted MNIST benchmark for out-of-distribution robustness.

MNIST-C applies 15 corruptions (noise, blur, affine transforms, digital and weather-like artifacts) to the MNIST train and test sets at a single severity per corruption. Each corruption is exposed as a builder config; the identity config is the uncorrupted MNIST reference shipped in the same archive.

Usage:

dataset = MNISTC(config_name=”shot_noise”, split=”test”) dataset = MNISTC(config_name=”identity”, split=”train”)

BUILDER_CONFIGS: list = [BuilderConfig(name='shot_noise', version=None, description='Poisson (shot) noise'), BuilderConfig(name='impulse_noise', version=None, description='Salt-and-pepper (impulse) noise'), BuilderConfig(name='glass_blur', version=None, description='Glass blur'), BuilderConfig(name='motion_blur', version=None, description='Motion blur'), BuilderConfig(name='shear', version=None, description='Affine shear'), BuilderConfig(name='scale', version=None, description='Affine scale'), BuilderConfig(name='rotate', version=None, description='Affine rotation'), BuilderConfig(name='brightness', version=None, description='Brightness increase'), BuilderConfig(name='translate', version=None, description='Affine translation'), BuilderConfig(name='stripe', version=None, description='Vertical stripe inversion'), BuilderConfig(name='fog', version=None, description='Fog'), BuilderConfig(name='spatter', version=None, description='Spatter'), BuilderConfig(name='dotted_line', version=None, description='Overlaid dotted line'), BuilderConfig(name='zigzag', version=None, description='Overlaid zigzag line'), BuilderConfig(name='canny_edges', version=None, description='Canny edge map'), BuilderConfig(name='identity', version=None, description='Uncorrupted MNIST reference')]

DEFAULT_CONFIG_NAME: str | None = 'shot_noise'

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/google-research/mnist-c', assets=mappingproxy({'train': DownloadInfo(url='https://zenodo.org/records/3239543/files/mnist_c.zip?download=1', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://zenodo.org/records/3239543/files/mnist_c.zip?download=1', fallbacks=[], checksum=None, filename=None)}), citation='@article{mu2019mnist,\n                        title={MNIST-C: A Robustness Benchmark for Computer Vision},\n                        author={Mu, Norman and Gilmer, Justin},\n                        journal={arXiv preprint arXiv:1906.02337},\n                        year={2019}}', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.mnist_m module

class MNISTM(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

MNIST-M dataset for digit classification and domain adaptation.

MNIST-M was introduced by Ganin & Lempitsky (2015) as the target domain for unsupervised domain adaptation from MNIST. Each image is an MNIST digit blended over a random color patch cropped from the BSDS500 photo collection, using the absolute difference between the digit and the patch. The result keeps MNIST’s labels but has natural-image color and texture statistics, so models trained only on MNIST transfer poorly to it.

This builder reads the original release by the authors (mnist_m.tar.gz, linked from Yaroslav Ganin’s homepage). It has 59001 training and 9001 test RGB images of size 32x32 in 10 classes (digits 0-9), stored as PNG files with one <file name> <label> labels file per split. Other copies circulating online, such as 60000/10000-image regenerations at 28x28, are not the same data.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://yaroslav.ganin.net/', assets=mappingproxy({'train': DownloadInfo(url='https://drive.usercontent.google.com/download?id=0B_tExHiYS-0veklUZHFYT19KYjg&export=download&confirm=t&resourcekey=0-DE7ivVVSn8PwP9fdHvSUfA', fallbacks=[], checksum='sha256:022d6584383f38fce6f53ca1387dcac9a51986cee5e4d231544546ec103501a6', filename='mnist_m.tar.gz'), 'test': DownloadInfo(url='https://drive.usercontent.google.com/download?id=0B_tExHiYS-0veklUZHFYT19KYjg&export=download&confirm=t&resourcekey=0-DE7ivVVSn8PwP9fdHvSUfA', fallbacks=[], checksum='sha256:022d6584383f38fce6f53ca1387dcac9a51986cee5e4d231544546ec103501a6', filename='mnist_m.tar.gz')}), citation='@inproceedings{ganin2015unsupervised,\n                    title     = {Unsupervised Domain Adaptation by Backpropagation},\n                    author    = {Ganin, Yaroslav and Lempitsky, Victor},\n                    booktitle = {Proceedings of the 32nd International Conference on Machine Learning},\n                    pages     = {1180--1189},\n                    year      = {2015}}', license='Unknown', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.not_mnist module

class NotMNIST(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

NotMNIST Dataset that contains images of letters A-J.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://yaroslavvb.blogspot.com/2011/09/notmnist-dataset.html', assets=mappingproxy({'train_images': DownloadInfo(url='https://github.com/davidflanagan/notMNIST-to-MNIST/raw/refs/heads/master/train-images-idx3-ubyte.gz', fallbacks=[], checksum=None, filename=None), 'train_labels': DownloadInfo(url='https://github.com/davidflanagan/notMNIST-to-MNIST/raw/refs/heads/master/train-labels-idx1-ubyte.gz', fallbacks=[], checksum=None, filename=None), 'test_images': DownloadInfo(url='https://github.com/davidflanagan/notMNIST-to-MNIST/raw/refs/heads/master/t10k-images-idx3-ubyte.gz', fallbacks=[], checksum=None, filename=None), 'test_labels': DownloadInfo(url='https://github.com/davidflanagan/notMNIST-to-MNIST/raw/refs/heads/master/t10k-labels-idx1-ubyte.gz', fallbacks=[], checksum=None, filename=None)}), citation='@misc{bulatov2011notmnist,\n                          author={Yaroslav Bulatov},\n                          title={notMNIST dataset},\n                          year={2011},\n                          url={http://yaroslavvb.blogspot.com/2011/09/notmnist-dataset.html}\n                        }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.office_home module

class OfficeHome(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Office-Home.

A domain adaptation / generalization benchmark with 15,588 images of 65 everyday office and home objects in 4 domains: art (2,427), clipart (4,365), product (4,439) and real world (4,357). The class set is shared across domains.

The release has no official split, so the configs define one:

"random" (default)

Stratified 80/20 train/test split inside every (domain, class) pair. Images are assigned by a SHA-256 of their bytes, so the split is reproducible.

"art", "clipart", "product", "real_world"

Leave-one-domain-out: train is every image of the other three domains and test is every image of the named domain.

Every example has image (RGB, variable size), label (65 classes) and domain (4 domains).

Usage:

train = OfficeHome(split=”train”) target_test = OfficeHome(config_name=”clipart”, split=”test”)

BUILDER_CONFIGS: list = [BuilderConfig(name='random', version=None, description='Stratified 80/20 train/test split within every (domain, class) pair.'), BuilderConfig(name='art', version=None, description="Leave-one-domain-out: train on the other domains, test on every image of 'art'."), BuilderConfig(name='clipart', version=None, description="Leave-one-domain-out: train on the other domains, test on every image of 'clipart'."), BuilderConfig(name='product', version=None, description="Leave-one-domain-out: train on the other domains, test on every image of 'product'."), BuilderConfig(name='real_world', version=None, description="Leave-one-domain-out: train on the other domains, test on every image of 'real_world'.")]

CLASS_NAMES = ('alarm_clock', 'backpack', 'batteries', 'bed', 'bike', 'bottle', 'bucket', 'calculator', 'calendar', 'candles', 'chair', 'clipboards', 'computer', 'couch', 'curtains', 'desk_lamp', 'drill', 'eraser', 'exit_sign', 'fan', 'file_cabinet', 'flipflops', 'flowers', 'folder', 'fork', 'glasses', 'hammer', 'helmet', 'kettle', 'keyboard', 'knives', 'lamp_shade', 'laptop', 'marker', 'monitor', 'mop', 'mouse', 'mug', 'notebook', 'oven', 'pan', 'paper_clip', 'pen', 'pencil', 'postit_notes', 'printer', 'push_pin', 'radio', 'refrigerator', 'ruler', 'scissors', 'screwdriver', 'shelf', 'sink', 'sneakers', 'soda', 'speaker', 'spoon', 'tv', 'table', 'telephone', 'toothbrush', 'toys', 'trash_can', 'webcam')

DEFAULT_CONFIG_NAME: str | None = 'random'

DOMAINS = ('art', 'clipart', 'product', 'real_world')

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.hemanthdv.org/officeHomeDataset.html', assets=mappingproxy({'archive': DownloadInfo(url='https://drive.usercontent.google.com/download?id=0B81rNlvomiwed0V1YUxQdC1uOTg&export=download&confirm=t&resourcekey=0-2SNWq0CDAuWOBRRBL7ZZsw', fallbacks=[], checksum=None, filename='OfficeHomeDataset_10072016.zip')}), citation='@inproceedings{venkateswara2017deep,\n            title={Deep Hashing Network for Unsupervised Domain Adaptation},\n            author={Venkateswara, Hemanth and Eusebio, Jose and Chakraborty, Shayok and Panchanathan, Sethuraman},\n            booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},\n            pages={5018--5027},\n            year={2017}}', license='Custom fair-use notice, non-commercial research and educational use only', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.olivetti_faces module

class OlivettiFaces(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Build the Olivetti Faces dataset for identity classification.

The dataset contains 400 grayscale 64x64 images of 40 individuals, with 10 images per individual.

This builder defines a deterministic split of 320 training images and 80 test images, using 8 training and 2 test images per identity. This split is not an official split provided by the original source.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://scikit-learn.org/stable/datasets/real_world.html#olivetti-faces-dataset', assets=mappingproxy({'train': DownloadInfo(url='https://ndownloader.figshare.com/files/5976027', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://ndownloader.figshare.com/files/5976027', fallbacks=[], checksum=None, filename=None)}), citation='@misc{olivettifaces,\n            title={The Olivetti Faces Dataset},\n            howpublished={AT&T Laboratories Cambridge}\n        }', license='Unknown', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.optdigits module

class Optdigits(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Build the UCI Optical Recognition of Handwritten Digits dataset.

The dataset contains 5,620 grayscale 8x8 handwritten digit samples across 10 classes.

It preserves the official UCI split of 3,823 training samples and 1,797 test samples. Original pixel intensities range from 0 to 16 and are converted to 8-bit grayscale images.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://archive.ics.uci.edu/dataset/80/optical+recognition+of+handwritten+digits', assets=mappingproxy({'train': DownloadInfo(url='https://archive.ics.uci.edu/ml/machine-learning-databases/optdigits/optdigits.tra', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://archive.ics.uci.edu/ml/machine-learning-databases/optdigits/optdigits.tes', fallbacks=[], checksum=None, filename=None)}), citation='@misc{uci_optdigits,\n            title={Optical Recognition of Handwritten Digits},\n            author={Alpaydin, E. and Kaynak, C.},\n            year={1998},\n            publisher={UCI Machine Learning Repository}\n        }', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.oracle_mnist module

class OracleMNIST(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Oracle-MNIST Dataset using raw IDX files for ancient character classification.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/wm-bupt/oracle-mnist', assets=mappingproxy({'train_images': DownloadInfo(url='https://raw.githubusercontent.com/wm-bupt/oracle-mnist/b6cb263f84d39a8bbb94c5b806d846b600068924/data/oracle/train-images-idx3-ubyte.gz', fallbacks=[], checksum='sha256:46f51f7565fea2f78628032dfa410c938d9a6bcd3691cade43419d143228bec6', filename=None), 'train_labels': DownloadInfo(url='https://raw.githubusercontent.com/wm-bupt/oracle-mnist/b6cb263f84d39a8bbb94c5b806d846b600068924/data/oracle/train-labels-idx1-ubyte.gz', fallbacks=[], checksum='sha256:45c65b236f8e169e6bd0875270d84cb132753351ca1938ae7935523ebd1e95fc', filename=None), 'test_images': DownloadInfo(url='https://raw.githubusercontent.com/wm-bupt/oracle-mnist/b6cb263f84d39a8bbb94c5b806d846b600068924/data/oracle/t10k-images-idx3-ubyte.gz', fallbacks=[], checksum='sha256:ca5e631c21fdbf452f2af8cb410d8546456b4b9820881b8d77150767496bf099', filename=None), 'test_labels': DownloadInfo(url='https://raw.githubusercontent.com/wm-bupt/oracle-mnist/b6cb263f84d39a8bbb94c5b806d846b600068924/data/oracle/t10k-labels-idx1-ubyte.gz', fallbacks=[], checksum='sha256:e10758deebac47b5ce0a1fe87cea09d50cb797544ed7b7dbac61e5f709c074b0', filename=None)}), citation='@article{wang2024dataset,\n                          title={A dataset of oracle characters for benchmarking machine learning algorithms},\n                          author={Wang, Mei and Deng, Weihong},\n                          journal={Scientific Data},\n                          volume={11},\n                          number={1},\n                          pages={87},\n                          year={2024}\n                        }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.oxford_iiit_pet module

class OxfordIIITPet(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Oxford-IIIT Pet Dataset

Abstract A 37 category pet dataset with roughly 200 images per class. Images have large variations in scale, pose, and lighting. Each image is annotated with a breed label (species + breed); this builder only consumes the breed label, not the trimap/bounding-box annotations.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.robots.ox.ac.uk/~vgg/data/pets/', assets=mappingproxy({'images': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/pets/data/images.tar.gz', fallbacks=[], checksum=None, filename=None), 'annotations': DownloadInfo(url='https://www.robots.ox.ac.uk/~vgg/data/pets/data/annotations.tar.gz', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{parkhi12a,\n                         title={Cats and Dogs},\n                         author={Parkhi, O. M. and Vedaldi, A. and Zisserman, A. and Jawahar, C. V.},\n                         booktitle={IEEE Conference on Computer Vision and Pattern Recognition},\n                         year={2012}}', license='CC BY-SA 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.pacs module

class PACS(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

PACS: Photo, Art painting, Cartoon and Sketch.

A domain generalization benchmark with 9,991 images of 7 object classes in 4 visual domains: photo (1,670), art painting (2,048), cartoon (2,344) and sketch (3,929).

The original release comes with an official train/validation split for every domain (<domain>_train_kfold.txt / <domain>_crossval_kfold.txt), which the authors ask to be used for fair comparisons. The configs follow it:

"all" (default)

train and validation are the official splits of all four domains.

"photo", "art_painting", "cartoon", "sketch"

Leave-one-domain-out: train and validation are the official splits of the other three domains, and test is every image of the named domain (<domain>_test_kfold.txt).

Every example has image (227x227 RGB), label (7 classes) and domain (4 domains).

Usage:

train = PACS(split=”train”) source_train = PACS(config_name=”sketch”, split=”train”) target_test = PACS(config_name=”sketch”, split=”test”)

BUILDER_CONFIGS: list = [BuilderConfig(name='all', version=None, description='Official train/validation splits of all four domains.'), BuilderConfig(name='photo', version=None, description="Leave-one-domain-out: train/validation are the official splits of the other domains, test is every image of 'photo'."), BuilderConfig(name='art_painting', version=None, description="Leave-one-domain-out: train/validation are the official splits of the other domains, test is every image of 'art_painting'."), BuilderConfig(name='cartoon', version=None, description="Leave-one-domain-out: train/validation are the official splits of the other domains, test is every image of 'cartoon'."), BuilderConfig(name='sketch', version=None, description="Leave-one-domain-out: train/validation are the official splits of the other domains, test is every image of 'sketch'.")]

CLASS_NAMES = ('dog', 'elephant', 'giraffe', 'guitar', 'horse', 'house', 'person')

DEFAULT_CONFIG_NAME: str | None = 'all'

DOMAINS = ('photo', 'art_painting', 'cartoon', 'sketch')

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://dali-dl.github.io/project_iccv2017.html', assets=mappingproxy({'images': DownloadInfo(url='https://drive.usercontent.google.com/download?id=0B6x7gtvErXgfbF9CSk53UkRxVzg&export=download&confirm=t&resourcekey=0-aQEyqs-mKSsoWHvclYvgjw', fallbacks=[], checksum=None, filename='PACS.zip'), 'art_painting_train_kfold': DownloadInfo(url='https://drive.usercontent.google.com/download?id=1cSo_5_cQLKoURsWwsKqFmw9neWFvDIeV&export=download&confirm=t', fallbacks=[], checksum=None, filename='pacs_art_painting_train_kfold.txt'), 'art_painting_crossval_kfold': DownloadInfo(url='https://drive.usercontent.google.com/download?id=1g7haGuvxmZA1SxPyGFZneDVi9wOm3nK4&export=download&confirm=t', fallbacks=[], checksum=None, filename='pacs_art_painting_crossval_kfold.txt'), 'art_painting_test_kfold': DownloadInfo(url='https://drive.usercontent.google.com/download?id=1KVe_S73jECKG9MsIujF-yqZAjPrf0y_O&export=download&confirm=t', fallbacks=[], checksum=None, filename='pacs_art_painting_test_kfold.txt'), 'cartoon_train_kfold': DownloadInfo(url='https://drive.usercontent.google.com/download?id=1EhlIyRoeZ5UPiUJs_UXGH-Ql1PS5qtYW&export=download&confirm=t', fallbacks=[], checksum=None, filename='pacs_cartoon_train_kfold.txt'), 'cartoon_crossval_kfold': DownloadInfo(url='https://drive.usercontent.google.com/download?id=1suUUrxyPOmN_dAVdhVSaG-NZDMOsl8JR&export=download&confirm=t', fallbacks=[], checksum=None, filename='pacs_cartoon_crossval_kfold.txt'), 'cartoon_test_kfold': DownloadInfo(url='https://drive.usercontent.google.com/download?id=14fpMKzcherr1onzp8jcyhmjJMjZSZ72h&export=download&confirm=t', fallbacks=[], checksum=None, filename='pacs_cartoon_test_kfold.txt'), 'photo_train_kfold': DownloadInfo(url='https://drive.usercontent.google.com/download?id=1RyCVv67zVITU4JYYr1XxLnG5q165yu8s&export=download&confirm=t', fallbacks=[], checksum=None, filename='pacs_photo_train_kfold.txt'), 'photo_crossval_kfold': DownloadInfo(url='https://drive.usercontent.google.com/download?id=1lXtOcPL-yWli_MuDZ0mQwLfq0XbzDu--&export=download&confirm=t', fallbacks=[], checksum=None, filename='pacs_photo_crossval_kfold.txt'), 'photo_test_kfold': DownloadInfo(url='https://drive.usercontent.google.com/download?id=1opW_IbwDQQmuylquX5UvRFxbj3muP5sz&export=download&confirm=t', fallbacks=[], checksum=None, filename='pacs_photo_test_kfold.txt'), 'sketch_train_kfold': DownloadInfo(url='https://drive.usercontent.google.com/download?id=13rWm64rW_dcdcyTFugu4fJcSCTkCB8YA&export=download&confirm=t', fallbacks=[], checksum=None, filename='pacs_sketch_train_kfold.txt'), 'sketch_crossval_kfold': DownloadInfo(url='https://drive.usercontent.google.com/download?id=1Rs0dUTduQ320bXwHBLQT1s82gyr92J9J&export=download&confirm=t', fallbacks=[], checksum=None, filename='pacs_sketch_crossval_kfold.txt'), 'sketch_test_kfold': DownloadInfo(url='https://drive.usercontent.google.com/download?id=1dFnnsWxYHvdCXmWdhPVwuJ8zgkb_wtbQ&export=download&confirm=t', fallbacks=[], checksum=None, filename='pacs_sketch_test_kfold.txt')}), citation='@inproceedings{li2017deeper,\n            title={Deeper, Broader and Artier Domain Generalization},\n            author={Li, Da and Yang, Yongxin and Song, Yi-Zhe and Hospedales, Timothy M.},\n            booktitle={Proceedings of the IEEE International Conference on Computer Vision (ICCV)},\n            pages={5542--5550},\n            year={2017}}', license='Unknown (research use; see the original release)', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.patch_camelyon module

class PatchCamelyon(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

PatchCamelyon (PCam): metastasis detection in lymph node histopathology patches.

PCam (Veeling et al., 2018) contains 327,680 96x96 RGB patches extracted from the Camelyon16 whole-slide images. A patch is labeled positive if its center 32x32 region contains metastatic tumor tissue. The official splits come from disjoint sets of slides:

• train: 262,144 patches

• validation: 32,768 patches (valid in the release)

• test: 32,768 patches

Rows also keep the per-patch metadata of the release: source slide, coordinates and whether the patch contains tumor outside the center region.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/basveeling/pcam', assets=mappingproxy({'train_x': DownloadInfo(url='https://zenodo.org/records/2546921/files/camelyonpatch_level_2_split_train_x.h5.gz', fallbacks=[], checksum='md5:1571f514728f59376b705fc836ff4b63', filename=None), 'train_y': DownloadInfo(url='https://zenodo.org/records/2546921/files/camelyonpatch_level_2_split_train_y.h5.gz', fallbacks=[], checksum='md5:35c2d7259d906cfc8143347bb8e05be7', filename=None), 'train_meta': DownloadInfo(url='https://zenodo.org/records/2546921/files/camelyonpatch_level_2_split_train_meta.csv', fallbacks=[], checksum='md5:5a3dd671e465cfd74b5b822125e65b0a', filename=None), 'validation_x': DownloadInfo(url='https://zenodo.org/records/2546921/files/camelyonpatch_level_2_split_valid_x.h5.gz', fallbacks=[], checksum='md5:d5b63470df7cfa627aeec8b9dc0c066e', filename=None), 'validation_y': DownloadInfo(url='https://zenodo.org/records/2546921/files/camelyonpatch_level_2_split_valid_y.h5.gz', fallbacks=[], checksum='md5:2b85f58b927af9964a4c15b8f7e8f179', filename=None), 'validation_meta': DownloadInfo(url='https://zenodo.org/records/2546921/files/camelyonpatch_level_2_split_valid_meta.csv', fallbacks=[], checksum='md5:67589e00a4a37ec317f2d1932c7502ca', filename=None), 'test_x': DownloadInfo(url='https://zenodo.org/records/2546921/files/camelyonpatch_level_2_split_test_x.h5.gz', fallbacks=[], checksum='md5:d8c2d60d490dbd479f8199bdfa0cf6ec', filename=None), 'test_y': DownloadInfo(url='https://zenodo.org/records/2546921/files/camelyonpatch_level_2_split_test_y.h5.gz', fallbacks=[], checksum='md5:60a7035772fbdb7f34eb86d4420cf66a', filename=None), 'test_meta': DownloadInfo(url='https://zenodo.org/records/2546921/files/camelyonpatch_level_2_split_test_meta.csv', fallbacks=[], checksum='md5:3455fd69135b66734e1008f3af684566', filename=None)}), citation='@article{veeling2018rotation,\n                        title={Rotation Equivariant {CNNs} for Digital Pathology},\n                        author={Veeling, Bastiaan S and Linmans, Jasper and Winkens, Jim and Cohen, Taco and Welling, Max},\n                        journal={arXiv preprint arXiv:1806.03962},\n                        year={2018}}\n                    @article{bejnordi2017diagnostic,\n                        title={Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer},\n                        author={Ehteshami Bejnordi, Babak and others},\n                        journal={JAMA},\n                        volume={318},\n                        number={22},\n                        pages={2199--2210},\n                        year={2017},\n                        doi={10.1001/jama.2017.14585}}', license='CC0 1.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.patternnet module

class PatternNet(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

PatternNet Remote Sensing Image Classification Dataset.

PatternNet is a large-scale high-resolution remote sensing dataset collected for performance evaluation of remote sensing image retrieval and scene classification. Images are collected from Google Earth or the Google Maps API.

The dataset contains 38 classes with 800 images per class (30,400 images total). Each image is 256x256 pixels in RGB format.

PatternNet has no official split. This builder splits each class by the number in its file names (e.g. airplane001.jpg):

• train: images 001-640 of each class, 24,320 images

• test: images 641-800 of each class, 6,080 images

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://sites.google.com/view/zhouwx/dataset', assets=mappingproxy({'train': DownloadInfo(url='https://nuisteducn1-my.sharepoint.com/:u:/g/personal/zhouwx_nuist_edu_cn/EYSPYqBztbBBqS27B7uM_mEB3R9maNJze8M1Qg9Q6cnPBQ?e=MSf977&download=1', fallbacks=[], checksum='sha256:456dd031950b0429518b8ec7d30a5c4b3f6456261ee1dd573a3eee43efb29958', filename='PatternNet.zip'), 'test': DownloadInfo(url='https://nuisteducn1-my.sharepoint.com/:u:/g/personal/zhouwx_nuist_edu_cn/EYSPYqBztbBBqS27B7uM_mEB3R9maNJze8M1Qg9Q6cnPBQ?e=MSf977&download=1', fallbacks=[], checksum='sha256:456dd031950b0429518b8ec7d30a5c4b3f6456261ee1dd573a3eee43efb29958', filename='PatternNet.zip')}), citation='@article{zhou2018patternnet,\n            title={PatternNet: A benchmark dataset for performance evaluation of remote sensing image retrieval},\n            author={Zhou, Weixun and Newsam, Shawn and Li, Congmin and Shao, Zhenfeng},\n            journal={ISPRS Journal of Photogrammetry and Remote Sensing},\n            volume={145},\n            pages={197--209},\n            year={2018},\n            publisher={Elsevier}\n        }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.places365_small module

class Places365Small(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

The Places365-Standard dataset (small version) for image classification.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='http://places2.csail.mit.edu/', assets=mappingproxy({'train': DownloadInfo(url='http://data.csail.mit.edu/places/places365/train_256_places365standard.tar', fallbacks=[], checksum=None, filename=None), 'val': DownloadInfo(url='http://data.csail.mit.edu/places/places365/val_256.tar', fallbacks=[], checksum=None, filename=None), 'devkit': DownloadInfo(url='http://data.csail.mit.edu/places/places365/filelist_places365-standard.tar', fallbacks=[], checksum=None, filename=None)}), citation='@article{zhou2017places,\n                         title={Places: A 10 million Image Database for Scene Recognition},\n                         author={Zhou, Bolei and Lapedriza, Agata and Khosla, Aditya and Oliva, Aude and Torralba, Antonio},\n                         year={2017}}\n            ', license='', checksums=None)

VERSION: Version = Version('1.0.0')

static extract_train_class(input_string)

Return the class path of a training image, e.g. "/a/airfield" for "data_256/a/airfield/00000001.jpg".

stable_datasets.images.plant_village module

class PlantVillage(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Load PlantVillage color images using the authors’ official train/test manifests.

BUILDER_CONFIGS: list = [BuilderConfig(name='color', version=None, description='')]

DEFAULT_CONFIG_NAME: str | None = 'color'

IMAGE_ARCHIVE = DownloadInfo(url='https://github.com/spMohanty/PlantVillage-Dataset/archive/refs/heads/master.zip', fallbacks=[], checksum=None, filename=None)

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/spMohanty/PlantVillage-Dataset', assets=mappingproxy({'train': DownloadInfo(url='https://huggingface.co/datasets/mohanty/PlantVillage/resolve/main/splits/color_train.txt', fallbacks=[], checksum=None, filename='plant-village-color-train.txt'), 'test': DownloadInfo(url='https://huggingface.co/datasets/mohanty/PlantVillage/resolve/main/splits/color_test.txt', fallbacks=[], checksum=None, filename='plant-village-color-test.txt')}), citation='@article{Mohanty2016PlantVillage,\n        title={Using Deep Learning for Image-Based Plant Disease Detection},\n        author={Mohanty, Sharada P. and Hughes, David P. and Salathe, Marcel},\n        journal={Frontiers in Plant Science},\n        volume={7},\n        year={2016},\n        doi={10.3389/fpls.2016.01419}\n        }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.plantdoc module

class PlantDoc(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

PlantDoc: plant disease classification from leaf images taken in the wild.

The classification data of the PlantDoc repository: 2,578 images of 13 plant species, one label per image, in 28 classes that are either a healthy leaf of a species or one of its diseases or pests. Unlike lab datasets such as PlantVillage, the images were collected from the internet and mostly show leaves in natural conditions, with cluttered backgrounds. The repository README calls this data Cropped-PlantDoc, but the images are the whole photos, not the 9,216 bounding-box crops that the paper calls Cropped-PlantDoc.

The official split has 2,342 train and 236 test images. “Tomato two spotted spider mites leaf” only has 2 train images and does not appear in the test split.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/pratikkayal/PlantDoc-Dataset', assets=mappingproxy({'train': DownloadInfo(url='https://github.com/pratikkayal/PlantDoc-Dataset/archive/5467f6012d78d1c446145d5f582da6096f852ae8.zip', fallbacks=[], checksum=None, filename='PlantDoc-Dataset.zip'), 'test': DownloadInfo(url='https://github.com/pratikkayal/PlantDoc-Dataset/archive/5467f6012d78d1c446145d5f582da6096f852ae8.zip', fallbacks=[], checksum=None, filename='PlantDoc-Dataset.zip')}), citation='@inproceedings{singh2020plantdoc,\n                         title={PlantDoc: A Dataset for Visual Plant Disease Detection},\n                         author={Singh, Davinder and Jain, Naman and Jain, Pranjali and Kayal, Pratik and\n                                 Kumawat, Sudhakar and Batra, Nipun},\n                         booktitle={Proceedings of the 7th ACM IKDD CoDS and 25th COMAD},\n                         pages={249--253},\n                         year={2020},\n                         doi={10.1145/3371158.3371196}}', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.rendered_sst2 module

class RenderedSST2(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Stanford Sentiment Treebank v2 sentences rendered as images for binary sentiment classification.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/openai/CLIP/blob/main/data/rendered-sst2.md', assets=mappingproxy({'train': DownloadInfo(url='https://openaipublic.azureedge.net/clip/data/rendered-sst2.tgz', fallbacks=[], checksum='sha256:cae70e31ca3dd38b5b44046e112276d7a9b4b6761b067b21decfd0b5b6777d05', filename=None), 'validation': DownloadInfo(url='https://openaipublic.azureedge.net/clip/data/rendered-sst2.tgz', fallbacks=[], checksum='sha256:cae70e31ca3dd38b5b44046e112276d7a9b4b6761b067b21decfd0b5b6777d05', filename=None), 'test': DownloadInfo(url='https://openaipublic.azureedge.net/clip/data/rendered-sst2.tgz', fallbacks=[], checksum='sha256:cae70e31ca3dd38b5b44046e112276d7a9b4b6761b067b21decfd0b5b6777d05', filename=None)}), citation='@inproceedings{radford2021learning,\n    title={Learning Transferable Visual Models From Natural Language Supervision},\n    author={Radford, Alec and Kim, Jong Wook and Hallacy, Chris and Ramesh, Aditya and Goh, Gabriel and Agarwal, Sandhini and Sastry, Girish and Askell, Amanda and Mishkin, Pamela and Clark, Jack and Krueger, Gretchen and Sutskever, Ilya},\n    booktitle={International Conference on Machine Learning (ICML)},\n    year={2021}}\n@inproceedings{socher2013recursive,\n    title={Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank},\n    author={Socher, Richard and Perelygin, Alex and Wu, Jean and Chuang, Jason and Manning, Christopher D. and Ng, Andrew and Potts, Christopher},\n    booktitle={Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing (EMNLP)},\n    year={2013}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.rock_paper_scissor module

class RockPaperScissor(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Rock Paper Scissors dataset.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://laurencemoroney.com/datasets.html', assets=mappingproxy({'train': DownloadInfo(url='https://storage.googleapis.com/download.tensorflow.org/data/rps.zip', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://storage.googleapis.com/download.tensorflow.org/data/rps-test-set.zip', fallbacks=[], checksum=None, filename=None)}), citation='@misc{laurence2019rock,\n                         title={Rock Paper Scissors Dataset},\n                         author={Laurence Moroney},\n                         year={2019},\n                         url={https://laurencemoroney.com/datasets.html}}', license='CC By 2.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.scene8 module

Original Spatial Envelope eight-category outdoor scene collection.

class Scene8(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Spatial Envelope outdoor scene classification dataset.

Contains 2,688 RGB images at 256 x 256 pixels across eight categories: coast, forest, highway, insidecity, mountain, opencountry, street, and tallbuilding. The original collection has no official partition; all images are exposed as train. Published by Oliva and Torralba (2001).

CLASS_NAMES = ('coast', 'forest', 'highway', 'insidecity', 'mountain', 'opencountry', 'street', 'tallbuilding')

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://people.csail.mit.edu/torralba/code/spatialenvelope/', assets=mappingproxy({'train': DownloadInfo(url='https://people.csail.mit.edu/torralba/code/spatialenvelope/spatial_envelope_256x256_static_8outdoorcategories.zip', fallbacks=[], checksum='sha256:e6157352dd9aeb055519be6afe1744bd20be9b2270bcb987f9dd06afb787f460', filename=None)}), citation='@article{oliva2001modeling,\n            title={Modeling the shape of the scene: A holistic representation of the spatial envelope},\n            author={Oliva, Aude and Torralba, Antonio},\n            journal={International Journal of Computer Vision}, volume={42}, pages={145--175}, year={2001}}\n        ', license='Not specified on the original dataset page; consult the publisher.', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.semeion module

class Semeion(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Semeion handwritten digit classification dataset.

Contains 1,593 grayscale images of handwritten digits (0-9). Each image is 16x16 pixels, represented by 256 binary pixel values in the original dataset.

The original dataset has no predefined train/test split. This implementation uses NumPy’s random generator with seed 42 to create a reproducible 80/20 split: - train: 1,274 images - test: 319 images

Each example returns a PIL grayscale image and an integer label between 0 and 9.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://archive.ics.uci.edu/dataset/178/semeion+handwritten+digit', assets=mappingproxy({'train': DownloadInfo(url='https://archive.ics.uci.edu/ml/machine-learning-databases/semeion/semeion.data', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://archive.ics.uci.edu/ml/machine-learning-databases/semeion/semeion.data', fallbacks=[], checksum=None, filename=None)}), citation='@misc{semeion_handwritten_digit_178,\n        title        = {{Semeion Handwritten Digit}},\n        year         = {1998},\n        howpublished = {UCI Machine Learning Repository},\n        note         = {{DOI}: https://doi.org/10.24432/C5SC8V}\n        }', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.shapes3d module

class Shapes3D(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Shapes3D dataset: 10x10x10x8x4x15 factor combinations, 64x64 RGB images.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://github.com/google-deepmind/3dshapes-dataset/', assets=mappingproxy({'train': DownloadInfo(url='https://huggingface.co/datasets/randall-lab/shapes3d/resolve/main/shapes3d.npz', fallbacks=[], checksum=None, filename=None)}), citation='@InProceedings{pmlr-v80-kim18b,\n  title = {Disentangling by Factorising},\n  author = {Kim, Hyunjik and Mnih, Andriy},\n  booktitle = {Proceedings of the 35th International Conference on Machine Learning},\n  pages = {2649--2658},\n  year = {2018},\n  editor = {Dy, Jennifer and Krause, Andreas},\n  volume = {80},\n  series = {Proceedings of Machine Learning Research},\n  month = {10--15 Jul},\n  publisher = {PMLR},\n  pdf = {http://proceedings.mlr.press/v80/kim18b/kim18b.pdf},\n  url = {https://proceedings.mlr.press/v80/kim18b.html}\n}', license='apache-2.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.small_norb module

class SmallNORB(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

SmallNORB dataset: 96x96 stereo images with 5 known factors.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://cs.nyu.edu/~ylclab/data/norb-v1.0-small/', assets=mappingproxy({'train': DownloadInfo(url='https://huggingface.co/datasets/randall-lab/small-norb/resolve/main/smallnorb-train.zip', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://huggingface.co/datasets/randall-lab/small-norb/resolve/main/smallnorb-test.zip', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{lecun2004learning,\n  title={Learning methods for generic object recognition with invariance to pose and lighting},\n  author={LeCun, Yann and Huang, Fu Jie and Bottou, Leon},\n  booktitle={Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2004. CVPR 2004.},\n  volume={2},\n  pages={II--104},\n  year={2004},\n  organization={IEEE}\n}', license='Apache-2.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.stanford40 module

Stanford 40 Actions with the official suggested train/test split.

class Stanford40(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

40 human actions in still images: 4,000 training and 5,532 test images.

Images retain their full frame and original resolution. Bounding boxes are not used. Resize and normalize at training time. Class IDs follow the alphabetical order of the original action names, including spelling and capitalization. No additional validation split is created. All images are RGB: the few non-RGB source files (e.g. the grayscale feeding_a_horse_191.jpg) are converted to RGB.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='http://vision.stanford.edu/Datasets/40actions.html', assets=mappingproxy({'images': DownloadInfo(url='http://vision.stanford.edu/Datasets/Stanford40_JPEGImages.zip', fallbacks=['https://web.archive.org/web/20240804122919id_/http://vision.stanford.edu/Datasets/Stanford40_JPEGImages.zip'], checksum='sha256:a1a08cb27345e33680f3a33f9cf2214be4c0989b9b3365692ca5903b733ba2a9', filename=None), 'splits': DownloadInfo(url='http://vision.stanford.edu/Datasets/Stanford40_ImageSplits.zip', fallbacks=['https://web.archive.org/web/20240804115240id_/http://vision.stanford.edu/Datasets/Stanford40_ImageSplits.zip'], checksum='sha256:cc04534f632fd613ffeaff716018baa52ff471959c33179ef5d41bf6360474e5', filename=None)}), citation='@inproceedings{yao2011human,\n  title={Human Action Recognition by Learning Bases of Action Attributes and Parts},\n  author={Yao, Bangpeng and Jiang, Xiaoye and Khosla, Aditya and Lin, Andy Lai\n          and Guibas, Leonidas and Fei-Fei, Li},\n  booktitle={2011 International Conference on Computer Vision},\n  year={2011}\n}', license='Unknown', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.stanford_dogs module

class StanfordDogs(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Fine-grained image classification. The Stanford Dogs dataset contains 20,580 RGB images of 120 dog breeds, built from ImageNet. The official split has 100 training images per class (12,000 total) and the remaining 8,580 images for testing.

LISTS_DOWNLOAD = DownloadInfo(url='http://vision.stanford.edu/aditya86/ImageNetDogs/lists.tar', fallbacks=['https://web.archive.org/web/20241221112710id_/http://vision.stanford.edu/aditya86/ImageNetDogs/lists.tar'], checksum='sha256:34b47cacd9a98b5d150e084f24d29391c084c55272295ec65c85651bc35f4d6c', filename=None)

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='http://vision.stanford.edu/aditya86/ImageNetDogs/', assets=mappingproxy({'train': DownloadInfo(url='http://vision.stanford.edu/aditya86/ImageNetDogs/images.tar', fallbacks=['https://web.archive.org/web/20250724000413id_/http://vision.stanford.edu/aditya86/ImageNetDogs/images.tar'], checksum='sha256:a443fc2f9a851aa1a7b8e4fa4af94b8459fce2c9f5a2f4dd4669cf9673a4ec58', filename=None), 'test': DownloadInfo(url='http://vision.stanford.edu/aditya86/ImageNetDogs/images.tar', fallbacks=['https://web.archive.org/web/20250724000413id_/http://vision.stanford.edu/aditya86/ImageNetDogs/images.tar'], checksum='sha256:a443fc2f9a851aa1a7b8e4fa4af94b8459fce2c9f5a2f4dd4669cf9673a4ec58', filename=None)}), citation='@inproceedings{khosla2011novel,\n                         author    = {Aditya Khosla and Nityananda Jayadevaprakash and Bangpeng Yao and Li Fei-Fei},\n                         title     = {Novel Dataset for Fine-Grained Image Categorization},\n                         booktitle = {First Workshop on Fine-Grained Visual Categorization, IEEE Conference on Computer Vision and Pattern Recognition},\n                         year      = {2011},\n                         month     = {June},\n                         address   = {Colorado Springs, CO}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.stl10 module

class STL10(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

STL-10 Dataset

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://cs.stanford.edu/~acoates/stl10/', assets=mappingproxy({'train': DownloadInfo(url='https://cs.stanford.edu/~acoates/stl10/stl10_binary.tar.gz', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://cs.stanford.edu/~acoates/stl10/stl10_binary.tar.gz', fallbacks=[], checksum=None, filename=None), 'unlabeled': DownloadInfo(url='https://cs.stanford.edu/~acoates/stl10/stl10_binary.tar.gz', fallbacks=[], checksum=None, filename=None)}), citation='@article{coates2011analysis,\n                        title={An analysis of single-layer networks in unsupervised feature learning},\n                        author={Coates, Adam and Ng, Andrew Y},\n                        journal={AISTATS},\n                        year={2011}}', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.sun397 module

class SUN397(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

SUN-397

SUN-397 is a scene recognition dataset containing 108,754 images across 397 scene categories. The original dataset provides ten predefined train/test partitions, each containing 50 training images and 50 testing images per class. All ten partitions are included in this implementation. An additional all configuration provides all available images as a single training split, intended for self- supervised learning (SSL) pipelines that do not require predefined train/test partitions.

The original download link provided on the official SUN dataset website (https://3dvision.princeton.edu/projects/2010/SUN/) is no longer functional. Therefore, a third-party upload was used to obtain the dataset. To verify the integrity of the downloaded archive, its MD5 checksum was compared against the original checksum published on the official dataset website. The checksums matched, indicating that the downloaded archive matches the original distribution.

BUILDER_CONFIGS: list = [BuilderConfig(name='all', version=None, description='All SUN-397 images as one training set (no test split)'), BuilderConfig(name='partition_01', version=None, description='Partition (1) with 50 training and 50 testing images per class'), BuilderConfig(name='partition_02', version=None, description='Partition (2) with 50 training and 50 testing images per class'), BuilderConfig(name='partition_03', version=None, description='Partition (3) with 50 training and 50 testing images per class'), BuilderConfig(name='partition_04', version=None, description='Partition (4) with 50 training and 50 testing images per class'), BuilderConfig(name='partition_05', version=None, description='Partition (5) with 50 training and 50 testing images per class'), BuilderConfig(name='partition_06', version=None, description='Partition (6) with 50 training and 50 testing images per class'), BuilderConfig(name='partition_07', version=None, description='Partition (7) with 50 training and 50 testing images per class'), BuilderConfig(name='partition_08', version=None, description='Partition (8) with 50 training and 50 testing images per class'), BuilderConfig(name='partition_09', version=None, description='Partition (9) with 50 training and 50 testing images per class'), BuilderConfig(name='partition_10', version=None, description='Partition (10) with 50 training and 50 testing images per class')]

DEFAULT_CONFIG_NAME: str | None = 'all'

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://3dvision.princeton.edu/projects/2010/SUN/', assets=mappingproxy({'archive': DownloadInfo(url='https://huggingface.co/zhengli97/SUN397_Dataset/resolve/main/SUN397.tar.gz?download=true', fallbacks=[], checksum='md5:8ca2778205c41d23104230ba66911c7a', filename='SUN397.tar.gz'), 'partitions': DownloadInfo(url='https://3dvision.princeton.edu/projects/2010/SUN/download/Partitions.zip', fallbacks=[], checksum='md5:29a205c0a0129d21f36cbecfefe81881', filename='Partitions.zip')}), citation='@inproceedings{5539970,\n                    title     = {SUN database: Large-scale scene recognition from abbey to zoo},\n                    author    = {Xiao, Jianxiong and Hays, James and Ehinger, Krista A. and Oliva, Aude and Torralba, Antonio},\n                    year      = {2010},\n                    booktitle = {2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition},\n                    volume    = {},\n                    number    = {},\n                    pages     = {3485--3492},\n                    doi       = {10.1109/CVPR.2010.5539970}}', license='Non-commercial research use', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.svhn module

class SVHN(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

SVHN (Street View House Numbers) Dataset for image classification.

SVHN is a real-world image dataset for developing machine learning and object recognition algorithms with minimal requirement on data preprocessing and formatting. It can be seen as similar in flavor to MNIST, but incorporates an order of magnitude more labeled data (over 600,000 digit images) and comes from a significantly harder, unsolved, real world problem (recognizing digits and numbers in natural scene images). SVHN is obtained from house numbers in Google Street View images.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='http://ufldl.stanford.edu/housenumbers/', assets=mappingproxy({'train': DownloadInfo(url='http://ufldl.stanford.edu/housenumbers/train_32x32.mat', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='http://ufldl.stanford.edu/housenumbers/test_32x32.mat', fallbacks=[], checksum=None, filename=None), 'extra': DownloadInfo(url='http://ufldl.stanford.edu/housenumbers/extra_32x32.mat', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{netzer2011reading,\n                          title={Reading digits in natural images with unsupervised feature learning},\n                          author={Netzer, Yuval and Wang, Tao and Coates, Adam and Bissacco, Alessandro and Wu, Baolin and Ng, Andrew Y and others},\n                          booktitle={NIPS workshop on deep learning and unsupervised feature learning},\n                          volume={2011},\n                          number={2},\n                          pages={4},\n                          year={2011},\n                          organization={Granada}\n                        }', license='', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.tf_flowers module

TensorFlow Flowers with original image labels and photographer attribution.

class TFFlowers(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Load all 3,670 released photographs in the unsplit train container.

Original labels and duplicate records are retained. The exact photographer attribution line accompanies each sample. The original LICENSE.txt remains at flower_photos/LICENSE.txt inside the downloaded archive; its opening declaration is also available as dataset.info.license. Evaluation grouping and splits are the caller’s responsibility.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.tensorflow.org/datasets/catalog/tf_flowers', assets=mappingproxy({'train': DownloadInfo(url='https://storage.googleapis.com/download.tensorflow.org/example_images/flower_photos.tgz', fallbacks=[], checksum='sha256:4c54ace7911aaffe13a365c34f650e71dd5bf1be0a58b464e5a7183e3e595d9c', filename=None)}), citation="TensorFlow, TensorFlow Flowers (tf_flowers). https://www.tensorflow.org/datasets/catalog/tf_flowers. Individual photographers are credited in the archive's LICENSE.txt.", license='All images in this archive are licensed under the Creative Commons By-Attribution License, available at:\nhttps://creativecommons.org/licenses/by/2.0/\nThe photographers are listed below, thanks to all of them for making their work available, and please be sure to credit them for any use as per the license.', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.tiny_imagenet module

class TinyImagenet(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Tiny ImageNet dataset for image classification tasks. It contains 200 classes with 500 training images, 50 validation images, and 50 test images per class.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.kaggle.com/c/tiny-imagenet', assets=mappingproxy({'train': DownloadInfo(url='http://cs231n.stanford.edu/tiny-imagenet-200.zip', fallbacks=[], checksum=None, filename=None), 'validation': DownloadInfo(url='http://cs231n.stanford.edu/tiny-imagenet-200.zip', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='http://cs231n.stanford.edu/tiny-imagenet-200.zip', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{Le2015TinyIV,\n                          title={Tiny ImageNet Visual Recognition Challenge},\n                          author={Ya Le and Xuan S. Yang},\n                          year={2015}\n                        }', license='MIT License', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.tiny_imagenet_c module

class TinyImagenetC(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Tiny ImageNet-C dataset for image classification tasks with corruptions applied.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://zenodo.org/records/2536630', assets=mappingproxy({'test': DownloadInfo(url='https://zenodo.org/records/2536630/files/Tiny-ImageNet-C.tar?download=1', fallbacks=[], checksum=None, filename=None)}), citation='@article{hendrycks2019robustness,\n                        title={Benchmarking Neural Network Robustness to Common Corruptions and Perturbations},\n                        author={Dan Hendrycks and Thomas Dietterich},\n                        journal={Proceedings of the International Conference on Learning Representations},\n                        year={2019}}', license='CC BY 4.0', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.usps module

class USPS(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

USPS handwritten digits from the LIBSVM distribution.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/multiclass.html#usps', assets=mappingproxy({'train': DownloadInfo(url='https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/multiclass/usps.bz2', fallbacks=[], checksum=None, filename=None), 'test': DownloadInfo(url='https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/multiclass/usps.t.bz2', fallbacks=[], checksum=None, filename=None)}), citation='@article{hull1994database,\n            author={Jonathan J. Hull},\n            title={A Database for Handwritten Text Recognition Research},\n            journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},\n            volume={16},\n            number={5},\n            pages={550--554},\n            year={1994},\n            doi={10.1109/34.291440}\n        }', license='Unknown', checksums=None)

VERSION: Version = Version('1.0.0')

stable_datasets.images.weizmann_seg module

class WeizmannSegmentation(*args, split=None, processed_cache_dir=None, download_dir=None, storage_format=None, backend_kwargs=None, decode_video=None, **kwargs)

Bases: BaseDatasetBuilder

Unsupervised figure-ground segmentation (evaluation only). The Weizmann Segmentation Evaluation Database contains 200 images: 100 showing one salient object (split one_object) and 100 showing two (split two_objects). Each image has 1-3 human segmentations (always 3 for one_object). There is no training split and no class labels.

Segmentations are stored as a list of PNG bytes (one per annotator); use decode_mask to get a 2-D map with 0 = background, 1 = red object, 2 = blue object. In two_objects, labels 1 and 2 are not matched across annotators (the same object can be 1 in one annotation and 2 in another), so match objects by overlap before combining annotators.

SOURCE: DatasetSource | Mapping = DatasetSource(homepage='https://www.wisdom.weizmann.ac.il/~vision/Seg_Evaluation_DB/', assets=mappingproxy({'one_object': DownloadInfo(url='https://www.wisdom.weizmann.ac.il/~vision/Seg_Evaluation_DB/download/Weizmann_Seg_DB_1obj.ZIP', fallbacks=[], checksum=None, filename=None), 'two_objects': DownloadInfo(url='https://www.wisdom.weizmann.ac.il/~vision/Seg_Evaluation_DB/download/Weizmann_Seg_DB_2obj.zip', fallbacks=[], checksum=None, filename=None)}), citation='@inproceedings{alpert2007image,\n                         title={Image Segmentation by Probabilistic Bottom-Up Aggregation and Cue Integration},\n                         author={Alpert, Sharon and Galun, Meirav and Basri, Ronen and Brandt, Achi},\n                         booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},\n                         year={2007}}', license='Research use only (custom terms, see the dataset website)', checksums=None)

VERSION: Version = Version('1.0.0')

decode_mask(data: bytes) → ndarray

Decode PNG bytes produced by encode_mask back to a 2-D uint8 label map.

encode_mask(mask: ndarray) → bytes

Encode a 2-D label map (0 = background, 1/2 = objects) as lossless PNG bytes.

Module contents