Hi, thanks for releasing V²-SAM.
I am trying to reproduce the reported Single-Expert results on the Ego-Exo4D Relation v2 test split using the released Fusion checkpoint and the evaluation code in this repository.
My Ego-to-Exo result is 44.63, which is close to the reported 44.5. However, the Exo-to-Ego result is 49.05, which is about 1.75 points higher than the reported 47.3.
Could you please clarify:
1.Does fusion_exo2ego_full.pth correspond to the Single-Expert result reported in the paper?
2.Were the released checkpoint or evaluation code updated after the reported experiments?
3.Is there any additional preprocessing or configuration required for reproducing the reported 47.3 Exo-to-Ego result?
4.Is the expected result for the current released checkpoint approximately 49.0 IoU?
ps：my result：Load checkpoint from weights/fusion_exo2ego_full.pth
Iter(test) [11629/11629]
Mean IoU: 0.4905
Hi, thanks for releasing V²-SAM.
I am trying to reproduce the reported Single-Expert results on the Ego-Exo4D Relation v2 test split using the released Fusion checkpoint and the evaluation code in this repository.
My Ego-to-Exo result is 44.63, which is close to the reported 44.5. However, the Exo-to-Ego result is 49.05, which is about 1.75 points higher than the reported 47.3.
Could you please clarify:
1.Does fusion_exo2ego_full.pth correspond to the Single-Expert result reported in the paper?
2.Were the released checkpoint or evaluation code updated after the reported experiments?
3.Is there any additional preprocessing or configuration required for reproducing the reported 47.3 Exo-to-Ego result?
4.Is the expected result for the current released checkpoint approximately 49.0 IoU?
ps：my result：Load checkpoint from weights/fusion_exo2ego_full.pth
Iter(test) [11629/11629]
Mean IoU: 0.4905