# Baseline comparison runners

Each script sets `VIDEO_PATH`, benchmark manifests, and calls `scripts/run_all_metrics.sh`.

Uses bundled paths via `scripts/eval_common.sh` — videos default to `videos/<benchmark>/<method>/`.

**Once per machine**: `bash scripts/setup_deps.sh /path/to/evaluations`  
**Optional**: `env.local.sh` for TVA / InternVL weight paths only.

## One-liner

```bash
bash scripts/run_eval.sh crosstest ours    # same as baselines/run_ours_crosstest.sh
```

## Ours (AgentHOI)

| Script | Benchmark | Videos (default) |
|--------|-----------|------------------|
| `run_ours_crosstest.sh` | Cross-test | `videos/crosstest/ours/` |
| `run_ours_AC.sh` | AC | `videos/AC/ours/` |
| `run_ours_homa.sh` | HOMA | `videos/homa_benchmark/ours/` |

## Compared methods

| Method | Scripts |
|--------|---------|
| VACE | `run_vace_crosstest.sh`, `run_vace_AC.sh` |
| HuMo | `run_humo_AC.sh` |
| HOMA | `run_homa_AC.sh` |

Replace mp4 under the matching `videos/` subfolder, or `export VIDEO_PATH=...` before running.

Results land in `results/ours_crosstest_eval/` (Object-DINO, VBench, TVA, InternVL subfolders).

## Aggregate all methods

```bash
bash scripts/aggregate_results.sh
# -> results/eval_summary.xlsx
```