Hi,
I have finally gotten my model trained to a level where I am happy and started converting. I noticed some frames where not converted and saw that the faces where missing (using the manual tool). Then I used some hours to add the missing masks hoping those frames would then be converted, however no such luck. I didn't spend much time fixing the landmarks, but made sure the masks (Unet-dfl) where OK.
Unfortunately I had no luck converting with the new masks in the alignment file. When using manual tool all frames I want to convert now have a mask (Unet-dfl), but for some reason they are just not converted. I'm sure there is something very obvious I'm missing here, and really appreciate if someone can push me in the right direction!
The reason so many faces are missing is likely that the video is blurry and it''s also an animated movie (Frozen). I'm quite impressed with the results as I know that blurry videos are not the best place to start, but I wanted to surprise my daughter transforming her to a Disney princess so had to try.. After 1 million iterations @batch size 16 (villain) it's pretty good.
Any ideas on how I can get the last frames converted? What am I missing after adding the masks?