Bemfoo
Results
3
issues of
Bemfoo
I used this index in BFM 2009. It's really useful, thanks a lot. Recently, I turn to use BFM 2017, and it has a version with only face cropped(model2017-1_face12_nomouth.h5), which...
After a serires of normal epochs, grad loss jumped to 1.0 suddenly and keep 1.0. Is there any possible problem?
After training, I noticed that all latent codes are not less than 0, which might be caused by F.relu at the end encoder. I think it's not neccessary and harmful...