| --- |
| title: README |
| emoji: 🌱 |
| colorFrom: green |
| colorTo: blue |
| sdk: static |
| pinned: false |
| short_description: DataSeeds.AI organization card |
| --- |
| |
| # DataSeeds.AI |
|
|
| Rights-cleared image and video training data, produced to spec by a managed crowd of photographers and on-the-ground collectors. Part of Zedge. |
|
|
| ## What you will find here |
|
|
| - **DSD sample dataset**: 7,770 peer-ranked photographs with scene descriptions, object labels, segmentation masks and full EXIF. Apache 2.0. |
| - **Baselines**: LLaVA-OneVision and BLIP2 fine-tuned on DSD, with the training code on [GitHub](https://github.com/DataSeeds-ai/DSD-finetune-blip-llava). |
| - **Paper**: [Peer-Ranked Precision](https://huggingface.co/papers/2506.05673), on whether peer-ranked, aesthetically scored data fine-tunes better than scraped web data. |
| - **Coming**: an open egocentric video sample for Physical AI and robotics. |
|
|
| ## How the data is made |
|
|
| Every image and clip is captured by a known contributor under a consent agreement, ranked by the community, and documented for GDPR, CCPA and BIPA. Provenance travels with the file. |
|
|
| ## Need more than the sample? |
|
|
| Larger collections, custom production and annotation: [dataseeds.ai](https://www.dataseeds.ai/) or sales@dataseeds.ai. |
