---
dataset_info:
  features:
  - name: index
    dtype: int64
  - name: url
    dtype: string
  splits:
  - name: train
    num_bytes: 670955648
    num_examples: 9011219
  - name: test
    num_bytes: 9330608
    num_examples: 125436
  - name: validation
    num_bytes: 3095477
    num_examples: 41620
  download_size: 365783126
  dataset_size: 683381733
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
  - split: validation
    path: data/validation-*
---

# Dataset Card for Open Images Dataset

This dataset contains images from the Open Images dataset. It includes image URLs, split into training, validation, and test sets.

## Dataset Details

### Dataset Description

Open Images is a dataset of approximately 9 million URLs to images that have been annotated with image-level labels, bounding boxes, object segmentation masks, and visual relationships.

- **Curated by:** Google LLC
- **License:** 
  - Images: [CC BY 2.0 license](https://creativecommons.org/licenses/by/2.0/)

### Dataset Sources

- **Repository:** [Open Images Dataset](https://storage.googleapis.com/openimages/web/index.html)

## Uses

### Direct Use

This dataset can be used for various computer vision tasks including image classification, object detection, segmentation, and visual relationship detection.

### Out-of-Scope Use

The dataset should not be used for any malicious activities, and users should verify the license status of each image themselves.

## Dataset Structure

The dataset is structured into three splits: train, test, and validation. Each split contains a parquet file with image URLs.

## Dataset Creation

### Curation Rationale

The dataset was created to advance research in computer vision by providing a large and diverse set of images.

### Source Data

#### Data Collection and Processing

The images were collected from various sources across the web and are provided under the Creative Commons Attribution 2.0 license.

#### Who are the source data producers?

The images were sourced from various photographers and websites under the Creative Commons Attribution 2.0 license.

## Bias, Risks, and Limitations

The dataset may contain biases inherent in the data sources. Users should be aware of these biases and exercise caution in interpreting the results.

### Recommendations

Users should be aware of the biases, risks, and limitations of the dataset. It is recommended to review the dataset for any biases that may affect the results of your specific use case.

## Citation

**BibTeX:**

```bibtex
@misc{openimages,
  author = {Google LLC},
  title = {Open Images Dataset},
  year = {2020},
  howpublished = {\url{https://storage.googleapis.com/openimages/web/index.html}},
  note = {Accessed: 2020-09-01}
}