---
license: cc-by-nc-4.0
tags:
- video-object-segmentation
- video-segmentation
- evaluation
---

# FaVOS

FaVOS evaluation uses the first annotated frame as the prompt frame. Metrics are
computed on the remaining frames only.

Download the dataset:

```bash
huggingface-cli download FaVOSsubmission/FaVOS --repo-type dataset --local-dir data
unzip data/JPEGImages.zip -d data
unzip data/Annotations.zip -d data
```

Dataset files are available at:
https://huggingface.co/datasets/FaVOSsubmission/FaVOS/tree/main

Evaluate predictions against:

```text
data/Annotations/<video_id>/<frame>.png
```

Use `favos-20.txt` and `favos-40.txt` as the evaluation splits.