# Demo cases

`demo_cases.csv` lists two HOI inference examples with **local assets** under `assets/`.

## CSV columns

| Column | Description |
|--------|-------------|
| `videoid` | Output id (`case_a`, `case_b`) |
| `image` | Human reference image |
| `object` / `object_llava` | Object reference crops |
| `audio` | Driving audio (wav) |
| `prompt` | Text description |
| `dwpose` | Sparse pose pickle |
| `fps` / `draw_type` / `expand_ratio` | Pose fps, pose draw mode, object expand ratio |

## Assets per case

```
assets/case_a/
  human.png, object.png, object_llava.png
  audio.wav, dwpose.pkl
```

To prepare your own case, use `scripts/preprocess.py` (see the main `README.md`).

Paths in `demo_cases.csv` are **relative to `RELEASE_ROOT`**.