pytorch-CycleGAN-and-pix2pix How to train on a custom rectangular 4:3 800x600 image dataset on CycleGAN?

How to train on a custom rectangular 4:3 800x600 image dataset on CycleGAN?

Open sarmientoj24 opened this issue 5 years ago • 5 comments

Is it possible to train images and generate images with that resolution? I got about 1000 dataset.

Nov 15 '20 06:11 sarmientoj24

If you have a 24GB memory GPU, that's possible. or you can load your dataset in 800x600 resolution, then crop images to 600x450, 16GB should be fine.

Nov 15 '20 06:11 JunlinHan

for 800x600 resolution, what should be the batch_size for that? Also, how to train using custom dataset?

Nov 15 '20 07:11 sarmientoj24

custom

just use batch_size 1, do not try large batch_size at first. As usual, split two domains to trainA and trainB.

Nov 15 '20 07:11 JunlinHan

if 12G , its ok for load_size 600 and crop_size 512

Jan 19 '22 12:01 improgress

Not sure. You can also do crop_size 360 or something smaller.

Jan 20 '22 22:01 junyanz