Models & Raw results:
- Google Drive
- baidu yun, code: rcsn
VOS test configuration directory: _experiments/sat/test/
| MODEL | Pipeline | Dataset | J&F-Mean | J-Mean | J-Recall | J-Decay | F-Mean | F-Recall | F-Decaly | FPS@GTX2080Ti | Config. Filename |
|---|---|---|---|---|---|---|---|---|---|---|---|
| SAT-Res50-transfer-param[1] | StateAwareTracker | DAVIS2017_val | 0.699069 | 0.667666 | 0.773786 | 0.174652 | 0.730471 | 0.82405 | 0.199258 | ~35 | sat_res50-davis17.yaml |
Nota:
[1] Because the parameters of model are transferred from our internel training framework, the result is lower than what presented in the paper. We are working on rewrite the training process with pytorch. Please keep tuned for better models.