Note

This is a Hugging Face dataset. Learn how to load datasets from the Hub in the Hugging Face integration docs.

Hugging Face

Dataset Card for StanfordDogsImbalanced#

image/png

This is a FiftyOne dataset with 19060 samples.

Installation#

If you haven’t already, install FiftyOne:

pip install -U fiftyone

Usage#

import fiftyone as fo
import fiftyone.utils.huggingface as fouh

# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = fouh.load_from_hub("Voxel51/Stanford-Dogs-Imbalanced")

# Launch the App
session = fo.launch_app(dataset)

Dataset Details#

Dataset Description#

An imbalanced version of the Stanford Dogs dataset designed for testing class imbalance mitigation techniques, including but not limited to synthetic data generation.

This version of the dataset was constructed by randomly splitting the original dataset into train, val, and test sets with a 60/20/20 split. For 15 randomly chosen classes, we then removed all but 10 of the training examples.

# Split the dataset into train, val, and test sets
import fiftyone.utils.random as four
train, val, test = four.random_split(dataset, split_fracs=(0.6, 0.2, 0.2))
splits_dict = { "train": train, "val": val, "test": test }

# Get the classes to limit
import random
classes = list(dataset.distinct("ground_truth.label"))
classes_to_limit = random.sample(classes, 15)

# Limit the number of samples for the selected classes
for class_name in classes_to_limit:
    class_samples = dataset.match(F("ground_truth.label") == class_name)
    samples_to_keep = class_samples.take(10)
    samples_to_remove = class_samples.exclude(samples_to_keep)
    dataset.delete_samples(samples_to_remove)

  • Curated by: [More Information Needed]

  • Funded by [optional]: [More Information Needed]

  • Shared by [optional]: [More Information Needed]

  • Language(s) (NLP): en

  • License: [More Information Needed]

Dataset Sources#

  • Paper: [More Information Needed]

  • Homepage: [More Information Needed]

Uses#

  • Fine-grained visual classification

  • Class imbalance mitigation strategies

Dataset Structure#

The following classes only have 10 samples in the train split:

  • Australian_terrier

  • Saluki

  • Cardigan

  • standard_schnauzer

  • Eskimo_dog

  • American_Staffordshire_terrier

  • Lakeland_terrier

  • Lhasa

  • cocker_spaniel

  • Greater_Swiss_Mountain_dog

  • basenji

  • toy_terrier

  • Chihuahua

  • Walker_hound

  • Shih-Tzu

  • Newfoundland

Citation#

BibTeX:

@inproceedings{KhoslaYaoJayadevaprakashFeiFei_FGVC2011,
  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",
  2011,
  month = "June",
  address = "Colorado Springs, CO",
}

Dataset Card Author#

Jacob Marks

Dataset Contacts#

aditya86@cs.stanford.edu and bangpeng@cs.stanford.edu