gpic

<h1 align='center' style="text-align:center; font-weight:bold; font-size:2.0em;letter-spacing:2.0px;"> GPIC: A Giant Permissive Image Corpus for Visual Generation</h1> <p align='center' style="text-align:center;font-size:1.25em;"> <a href="https://keshik6.github.io/" target="_blank" style="text-decoration: none;">Keshigeyan&nbsp;Chandrasegaran</a>^*1,&nbsp; <a href="https://kylesargent.github.io/" target="_blank" style="text-decoration: none;">Kyle&nbsp;Sargent</a>^*1,&nbsp; <a href="https://www.suchiragarwal.com/" target="_blank" style="text-decoration: none;">Suchir&nbsp;Agarwal</a>¹,&nbsp; <a href="https://mjang01011.github.io/portfolio/" target="_blank" style="text-decoration: none;">Michael&nbsp;Jang</a>¹,&nbsp;<br/> <a href="https://zymrael.github.io/" target="_blank" style="text-decoration: none;">Michael&nbsp;Poli</a>^1,2,&nbsp; <a href="https://www.niebles.net/" target="_blank" style="text-decoration: none;">Juan&nbsp;Carlos&nbsp;Niebles</a>^1,4,&nbsp; <a href="https://cs.stanford.edu/people/jcjohns/" target="_blank" style="text-decoration: none;">Justin&nbsp;Johnson</a>³,&nbsp; <a href="https://jiajunwu.com" target="_blank" style="text-decoration: none;">Jiajun&nbsp;Wu</a>¹,&nbsp; <a href="https://profiles.stanford.edu/fei-fei-li" target="_blank" style="text-decoration: none;">Li&nbsp;Fei-Fei</a>¹<br/> <span class="author-block">¹&nbsp;Stanford University&nbsp;&nbsp;</span> <span class="author-block">²&nbsp;Radical Numerics&nbsp;&nbsp;</span> <span class="author-block">³&nbsp;University of Michigan&nbsp;&nbsp;</span> <span class="author-block">⁴&nbsp;Salesforce Research&nbsp;&nbsp;</span><br/> ^*&nbsp;Equal contribution<br/> <a href="https://arxiv.org/abs/2605.30341" title="arXiv" target="_blank" rel="nofollow" style="text-decoration: none;">📄 arXiv</a> | <a href="https://gpic.stanford.edu" title="Website" target="_blank" rel="nofollow" style="text-decoration: none;">🌎 Website</a> | <a href="https://huggingface.co/datasets/stanford-vision-lab/giant-permissive-image-corpus" title="Dataset" target="_blank" rel="nofollow" style="text-decoration: none;">🤗 Dataset</a> | <a href="https://huggingface.co/stanford-vision-lab/gpic-baselines" title="Models" target="_blank" rel="nofollow" style="text-decoration: none;">🤗 Models</a> | <a href="https://github.com/keshik6/gpic" title="Evaluation Toolkit" target="_blank" rel="nofollow" style="text-decoration: none;">🥇 Evaluation toolkit</a> </p> <p align="center"> <img src="https://huggingface.co/datasets/stanford-vision-lab/giant-permissive-image-corpus/resolve/main/figures/gpic_graphics.svg" alt="GPIC Dataset Overview" style="width: 100%; max-width: 3000px;"> </p> ## Abstract Studying scalable methods for visual generative modeling requires large, accessible, and stable datasets. We introduce **GPIC**, a **G**iant **P**ermissive **I**mage **C**orpus of approximately **28 trillion pixels**. GPIC comprises diverse internet images captioned by a state-of-the-art vision-language model, including 100M training, 200K validation, and 1M test examples. Moreover, all GPIC images are permissively licensed for both research and commercial use. GPIC is safety-filtered, deduplicated, and centrally hosted on Hugging Face. We provide a benchmarking protocol for generative modeling on GPIC. Finally, we provide a reference baseline for pixel-space flow matching on GPIC. Our dataset, benchmark, and models are available on [Hugging Face](https://huggingface.co/datasets/stanford-vision-lab/gpic). Evaluation toolkit and code are available at [gpic.stanford.edu](https://gpic.stanford.edu). We hope GPIC supports open, accessible, and reproducible research on large-scale visual generative modeling. ## GPIC Statistics <p align="center"> <img src="https://huggingface.co/datasets/stanford-vision-lab/giant-permissive-image-corpus/resolve/main/figures/gpic_stats.jpeg" alt="GPIC Stats Overview" style="width: 100%; max-width: 3000px;"> </p> ## Dataset Organization There are 8000 tars for GPIC train, 32 tars for validation, and 128 tars for test. ``` giant-permissive-image-corpus/ ├── train/ (8000 files, gpic_train_{00000–07999}.tar) ├── val/ (32 files, gpic_val_{00000–0031}.tar) ├── test/ (128 files, gpic_test_{00000–00127}.tar) ├── .gitattributes └── README.md ``` ### Tar File Format Each tar archive contains alternating image–metadata pairs: - `{key}.json` — metadata and caption - `{key}.jpg` / `{key}.png` — corresponding image Files are stored sequentially such that each JSON entry is followed by its corresponding image. For example: ``` {key_1}.json {key_1}.jpg {key_2}.json {key_2}.png {key_3}.json {key_3}.jpg ``` ### JSON Format Each json includes metadata in the following format: ``` { "retrieved_at": str, "license": str, "license_url": str, "attribution": str, "key": str, # unique identifier for the image "img_width": int, "img_height": int, "split": [str], # dataset split, one of {"nano", "lite", "full"} "caption_type": str, # one of {"tag", "short", "medium", "long"} "caption": str }, // Next image record ``` <!-- ## Limitations --> <!-- ## Usage Restrictions --> ## Contact For issues, feedback, or contributions, please open an issue or submit a pull request. - Keshigeyan Chandrasegaran: [e-mail] - Kyle Sargent: [e-mail] ### Image Removal Requests If you believe a specific GPIC image should be removed due to copyright, privacy, licensing, or safety concerns, please submit an image removal request using this form: [Image Removal Request](https://docs.google.com/forms/d/e/1FAIpQLSdd3un-uGf9z0gVNUtBFMCki-B3nKqmnEVvQRL_8QQh8J6uFw/viewform?usp=header). Please include the relevant GPIC image key(s). Requests are reviewed case by case by the dataset maintainers. ## Citation ```bibtex @misc{chandrasegaran2026gpic, title={GPIC: A Giant Permissive Image Corpus for Visual Generation}, author={Keshigeyan Chandrasegaran and Kyle Sargent and Suchir Agarwal and Michael Jang and Michael Poli and Juan Carlos Niebles and Justin Johnson and Jiajun Wu and Li Fei-Fei}, year={2026}, eprint={2605.30341}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2605.30341}, } ```