FaVOS

Project page | Paper

FaVOS is a video object segmentation benchmark with fractional temporal visibility. It contains 200 videos and 279 annotated objects that may disappear and reappear.

Subset	Videos	Objects	Mean τ
FaVOS-20	100	129	0.2051
FaVOS-40	100	150	0.4048
Full	200	279	0.3124

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

Download the dataset:

huggingface-cli download snu-aidas/FaVOS --repo-type dataset --local-dir data
unzip data/JPEGImages.zip -d data
unzip data/Annotations.zip -d data

Evaluate predictions against:

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

Use favos-20.txt and favos-40.txt as the evaluation splits. To compute Volumetric J&F with eval_jnf_v.py:

python data/eval_jnf_v.py --pred-root predictions --output-dir results/jf_v

License

The annotations and the 67 self-collected videos are released under CC BY-NC 4.0. Frames from the 133 YFCC100M videos keep their original licenses. See LICENSE.md.

Contact: csjihwanh@snu.ac.kr

Citation

@inproceedings{hong2026favos,
  title     = {When Predicting Nothing Beats SAM 3: Revisiting Evaluation in Video Object Segmentation},
  author    = {Hong, Jihwan and Park, Woohyeon and Kim, Jaeik and Do, Jaeyoung},
  booktitle = {Advances in Neural Information Processing Systems},
  year      = {2026},
  note      = {Evaluations \& Datasets Track}
}