This script is designed to compute the J and F scores where the annotations are sampled. In other words, the groundtruths are not provided for every frame.
python3 score_longvideo.py \
--path /path/to/longvideo_dataset \
--results_path ../output/AFB-URR_LongVideo \
--method AFB-URR \
--updateGlobal results for AFB-URR
| J&F-Mean | J-Mean | J-Recall | J-Decay | F-Mean | F-Recall | F-Decay |
|---|---|---|---|---|---|---|
| 0.832571 | 0.827035 | 0.916667 | 0.115116 | 0.838107 | 0.916667 | 0.138554 |
Per sequence results for AFB-URR
| Sequence | J-Mean | F-Mean |
|---|---|---|
| rat_1 | 0.745197 | 0.763893 |
| dressage_1 | 0.834371 | 0.874584 |
| blueboy_1 | 0.901536 | 0.875845 |
These codes are modified from DAVIS VOS Evaluation. If you use this code in your work, please cite both DAVIS dataset and our paper.