The result is biased

[haozhihui110]

Hello, when reproducing the FD-GAN code, use the command line given in your github, the only difference is batch_size = 16 instead of 256, the final result is only 70.9%, what is the reason for this, it is batch_size The value is too small? Or is there some other reason? Thank you

[yxgeee]

Yes, the batch size would affect the final performance, since the training of batch norm layers in the network is critical for optimal performance.