everanimate / data /README.md
vita-video-gen's picture
Add demo data and ablation videos
cd53bfa verified
# Minimal Data
This directory contains a tiny smoke-test dataset for checking training and inference wiring.
## Training
`train/metadata.csv` contains one training sample with the same columns expected by `examples/wanvideo/model_training/train_svi.py`:
```text
video,prompt,animate_pose_video,animate_face_video
```
The paths in the CSV are relative to `data/train/`.
The toy training videos are stored with normalized sample names:
```text
videos/sample_000001.mp4
poses/sample_000001_poses.mp4
faces/sample_000001_faces.mp4
```
## Inference
`test/` contains a single inference case:
```text
frames/demo_000001.png
poses/demo_000001_poses.mp4
faces/demo_000001_faces.mp4
```
This matches the default layout used by the top-level `inference.sh`.
## Ablation
`ablation/` contains the two-stage ablation outputs:
```text
ablation/stage1.mp4
ablation/stage2.mp4
```