I'm trying to train a PGAN with target image size of 1024x1024. It's currently training at size 64x64 and it's painfully slow to get through it with 48,000 iterations. Besides cutting down on the number of iterations, is there a faster way to train this model (e.g. fp16 format)? Thanks.
I'm trying to train a PGAN with target image size of 1024x1024. It's currently training at size 64x64 and it's painfully slow to get through it with 48,000 iterations. Besides cutting down on the number of iterations, is there a faster way to train this model (e.g. fp16 format)? Thanks.