Dataset Card for ImageNet-O

This is a FiftyOne dataset with 2000 samples.

The recipe notebook for creating this dataset can be found here.

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/ImageNet-O")

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

Dataset Details

Dataset Description

The ImageNet-O dataset consists of images from classes not found in the standard ImageNet-1k dataset. It tests the robustness and out-of-distribution detection capabilities of computer vision models trained on ImageNet-1k.

Key points about ImageNet-O:

• Contains images from classes distinct from the 1,000 classes in ImageNet-1k

• Enables testing model performance on out-of-distribution samples, i.e. images that are semantically different from the training data

• Commonly used to evaluate out-of-distribution detection methods for models trained on ImageNet

• Reported using the Area Under the Precision-Recall curve (AUPR) metric

• Manually annotated, naturally diverse class distribution, and large scale

• Curated by: Dan Hendrycks, Kevin Zhao, Steven Basart, Jacob Steinhardt, Dawn Song

• Shared by: Harpreet Sahota, Hacker-in-Residence at Voxel51

• Language(s) (NLP): en

• License: MIT License

Dataset Sources [optional]

• Repository: https://github.com/hendrycks/natural-adv-examples
• Paper: https://arxiv.org/abs/1907.07174

Citation

BibTeX:

@article{hendrycks2021nae,
  title={Natural Adversarial Examples},
  author={Dan Hendrycks and Kevin Zhao and Steven Basart and Jacob Steinhardt and Dawn Song},
  journal={CVPR},
  year={2021}
}