test on ubfc-phys gives bad results.
ycyoon opened this issue · comments
I tested PURE, UBFC2, COFACE, and UBFC-Phys with mtts-can model.
Only a model trained with ubfc-phys gives bad results. (overfitted on the training dataset and fails on valid set)
I used s50~s56 as a valid set and trained with other videos.
Any idea? or does Anyone succeed in training with the UBFC-Phys dataset?
For mtts-can we used pretrained models as indicated in the paper. We have not done training experiment so it is difficult for us to make a judgment on this.
best
pyVHR team
Why it showed really bad result when i use the pre-trained model to test UBFC videos? I just ran the demo_deep notebook. Do i need to do other things?