Ground truth data for EWMBench evaluation, extracted and restructured for direct use by the wm-evaluation-harness framework.
Extracted from agibot-world/EWMBench (gt_dataset.tar). Original dataset by AgibotTech.
{task_id}/
├── {episode_id}/
│ ├── prompt/
│ │ ├── init_frame.png # Initial frame for generation
│ │ └── prompt.txt # Task description
│ └── video/
│ ├──… See the full description on the dataset page:
https://huggingface.co/datasets/Physis-AI/wm-eval-gt-ewmbench.
