FaVOS

[NeurIPS 2026 E&D] Official Implementation for "When Predicting Nothing Beats SAM 3: Revisiting Evaluation in Video Object Segmentation"

Project page | Paper | Dataset

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

Evaluation Procedure

Download and extract 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

Install dependencies:

pip install -r requirements.txt

Prepare predictions with the same IDs and frame names as the annotations:

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

Run J&F_v evaluation:

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

The evaluator uses favos-20.txt and favos-40.txt automatically.

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}
}