Fruits-360

Overview

Fruits-360 is an image classification dataset of fruits, vegetables, nuts, seeds and mushrooms. Each object was filmed while rotating on a slow motor (3 rpm) in front of a white sheet of paper, and the background was then removed with a flood-fill algorithm and replaced with pure white. Frames are resized to 100×100 RGB. Different varieties of the same fruit (e.g. Apple Golden 1 and Apple Granny Smith 1) are separate classes, and the numeric suffix distinguishes different filming sessions of the same kind.

Train: 147,416 images
Test: 49,106 images
Classes: 274

The official Training/Test partitions are kept as released. Since the upstream repository is updated in place, the builder pins version 2026.10.18.2 (commit 18fbaa1). A few upstream class folders differ only in letter case (Blackberry 3 / BlackBerry 3 and Blackberry 4 / BlackBerry 4) and contain disjoint images of the same class; these are merged, giving the 274 classes reported by the authors.

Because consecutive frames of one rotating object are nearly identical and the train and test images come from the same videos, test accuracy on Fruits-360 is typically very high.

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`	100×100 RGB image of a single object on a white background
`label`	int	Class label (0-273)

Usage Example

Basic Usage

from stable_datasets.images.fruits360 import Fruits360

# First run will download + prepare cache, then return the split as a HF Dataset
ds_train = Fruits360(split="train")
ds_test = Fruits360(split="test")

# If you omit the split (split=None), you get a DatasetDict with all available splits
ds_all = Fruits360(split=None)

sample = ds_train[0]
print(sample.keys())  # {"image", "label"}
print(ds_train.info.features["label"].int2str(sample["label"]))  # e.g. "Almonds 1"

# Optional: make it PyTorch-friendly
ds_train_torch = ds_train.with_format("torch")

References

Homepage: https://github.com/fruits-360
Kaggle: https://www.kaggle.com/moltean/fruits
License: CC BY-SA 4.0

Citation

@article{muresan2018fruit,
  title={Fruit recognition from images using deep learning},
  author={Mure{\c{s}}an, Horea and Oltean, Mihai},
  journal={Acta Universitatis Sapientiae, Informatica},
  volume={10},
  number={1},
  pages={26--42},
  year={2018}
}