I have worked on BigGAN-deep codes from HuggingFace git repository before. I have the question of where you exactly applied PCA in the code? You mentioned in your paper "at the first linear layer of BigGAN512-deep". Could you please let me know if it was the output of Linear 4x4x16ch (in red box below from BigGAN-deep architecture)? If yes, the output of it would be a high dimensional vector with the size 32768 = 2048 x 4 x 4. Did you apply PCA on this high dimensional space? Thanks in advance for your reply!
I have worked on BigGAN-deep codes from HuggingFace git repository before. I have the question of where you exactly applied PCA in the code? You mentioned in your paper "at the first linear layer of BigGAN512-deep". Could you please let me know if it was the output of Linear 4x4x16ch (in red box below from BigGAN-deep architecture)? If yes, the output of it would be a high dimensional vector with the size 32768 = 2048 x 4 x 4. Did you apply PCA on this high dimensional space? Thanks in advance for your reply!