search

vincent-thevenin / Realistic-Neural-Talking-Head-Models

My implementation of Few-Shot Adversarial Learning of Realistic Neural Talking Head Models (Egor Zakharov et al.).
GNU General Public License v3.0
828 stars 195 forks source link

Bug in the code #66

Open phquanta opened 3 years ago

phquanta commented 3 years ago

I've started training and i've noticed that after reloading model checkpoint i'm seeing mismatch for W_i vectors with hardcoded dime1=32 parameter in save_disc branch and dim1=num_videos in master branch. None of them seems to be correct, looks like it should be batch_size because we are loading W_i at each training iteration.

hanweikung commented 1 year ago

I had the same problem. Thank you for the suggestion