Hi, thanks for your amazing project! Your paper and codebase really help me a lot.
But I found that you seem to forget to add the mask when the first 16 frames are empty in test_fvd_ddpm. That will cause the conditional value to be all -1 instead of all 0, thus leading to the prediction failure that the model tends to predict dark frames.
Hi, thanks for your amazing project! Your paper and codebase really help me a lot.
But I found that you seem to forget to add the mask when the first 16 frames are empty in
test_fvd_ddpm. That will cause the conditional value to be all -1 instead of all 0, thus leading to the prediction failure that the model tends to predict dark frames.