| license: other | |
| task_categories: | |
| - text-to-video | |
| - image-to-video | |
| tags: | |
| - visual-reasoning | |
| - world-engine | |
| - synthetic | |
| - tangram | |
| # Tangram WE-Task3 Dataset | |
| Synthetic 832x480, 81-frame Tangram solution videos generated by `we-task3`. | |
| - Train: 3,000 cases; easy, medium, and hard have 1,000 each. | |
| - IID validation: `validation/`; 25 cases per difficulty and 75 total. | |
| - Fixed OOD challenge: `ood/`; one original Task 3 example per difficulty. | |
| - IID cases exclude every train seed and canonical translation/global-rotation class. | |
| - See `SPLITS.md` and the split manifests for details. | |