Stanford Dogs

Overview

The Stanford Dogs dataset is designed for fine-grained image categorization. It contains 20,580 images of 120 dog breeds from around the world, built from ImageNet. The official split uses 100 images per class for training and the remaining images for testing.

• Train: 12,000 images (100 per class)

• Test: 8,580 images

Data Structure

When accessing an example using ds[i], you will receive a dictionary with the following keys:

Key	Type	Description
image	PIL.Image.Image	H×W×3 RGB image (sizes vary)
label	int	Class label (0-119), in the order of the official WordNet-ID folders (0 = Chihuahua, 119 = African_hunting_dog)

Notes

• Splits follow the official train_list.mat and test_list.mat files.

• Color mode: every image is returned as 3-channel RGB. The source archive contains one RGBA image (n02105855_2933.jpg), which is converted to RGB so that images can be batched together.

• Bounding boxes: the official per-image bounding-box annotations are not included; this builder provides class labels only.

• ImageNet overlap: all images come from ImageNet. Results from models pretrained on ImageNet are therefore not a clean transfer evaluation on this dataset.

• Download size: the image archive is about 794 MB.

Usage Example

Basic Usage

from stable_datasets.images.stanford_dogs import StanfordDogs

# First run will download + prepare cache, then return the split as a HF Dataset
ds = StanfordDogs(split="train")

# If you omit the split (split=None), you get a DatasetDict with all available splits
ds_all = StanfordDogs(split=None)

sample = ds[0]
print(sample.keys())  # {"image", "label"}

# Map a label back to its breed name
print(ds.info.features["label"].names[sample["label"]])

# Optional: make it PyTorch-friendly
ds_torch = ds.with_format("torch")

Related Datasets

• CUB-200: Fine-grained bird species classification

• FGVC Aircraft: Fine-grained aircraft model classification

• Cars196: Fine-grained car model classification

References

• Official website: http://vision.stanford.edu/aditya86/ImageNetDogs/

• License: not stated by the dataset authors; the images originate from ImageNet

Citation

@inproceedings{khosla2011novel,
author    = {Aditya Khosla and Nityananda Jayadevaprakash and Bangpeng Yao and Li Fei-Fei},
title     = {Novel Dataset for Fine-Grained Image Categorization},
booktitle = {First Workshop on Fine-Grained Visual Categorization, IEEE Conference on Computer Vision and Pattern Recognition},
year      = {2011},
month     = {June},
address   = {Colorado Springs, CO}
}