Question about classifier's batch size

[lmz123321]

Hi,sorry to bother you @lfz

I note that when training the classifier, you set b2=12 so batch_size for classifier=12

If we use 8 GPUs, that will be 2 image_patch for GPU 0-5 and 0 image_patch for GPU 6,7

Why do you set b2=12? If I set b2=16, Will the performance of classifier improve?

Thank your very much!