Junlin Han
Junlin Han
This one is unsupervised while pix2pix is supervised (.i.e, paired data for training).
~ 5000 iterations per epoch. I assume every iteration in CUT takes ~0.4 sec on a GTX 1080 GPU. (400 * 5000 * 0.4)/3600 = ~ 222 hours. FastCUT is...
Sorry dude, I can not fully understand your question. There's two datasets related to cat in this project. One is from AFHQ, around 5000 images in each domain ( cat...
> > Hi there thanks for your reply. I wanted to ask if the dataset doesn't need label, how it will be trained? How the network will understand how bad...
Did you use the correct command? This one: python train.py --model sincut --name singleimage_monet_etretat --dataroot ./datasets/single_image_monet_etretat It take some hours in a normal GPU (1080Ti or better one).
Emm, one simple solution is that you may create a testB with one image (one is enough).
Hi, You are correct! Just add --continue_train --epoch_count XX( where XX is the epoch in your automatically saved latest model)
Hi Hudaladi, No worries at all, I'm happy to help! 1: Do You think Your Project is capeable to take glass reflections away? Yes, BIDeN is OK to perform reflection...
From my experience, you probably need to train it for more epochs.
Make sure you have created trainA,trainB inside ./datasets/datasetchd, name them properly.