It is unstable using your new release code in 64*64 resolution

Question

It is unstable using your new release code in 64*64 resolution

akafen opened this issue 7 years ago · 6 comments

akafen commented 7 years ago

It is unstable using your newest release code in 64*64 resolution,why?

Answer 1 · 2017-12-06T06:29:46.000Z

Can't figure out why. In my experiment, when fading in 256x256, the model crashed.

Answer 2 · 2017-12-13T04:16:17.000Z

Try train_no_tanh.py with latest data.py script.

Answer 3 · 2017-12-21T17:52:35.000Z

yeah, same thing happened to me as well. In my case, the model collapsed very early on at 16*16 fade in. ( I used default value for all the hyper-parameters listed in the README)

--g_lr_max 1e-3 --d_lr_max 1e-3 --train_kimg 600 --transition_kimg 600 --beta1 0 --beta2 0.99 --gan lsgan --first_resol 4 --target_resol 256 --no_tanh

Answer 4 · 2017-12-22T08:48:41.000Z

Me too, the model coIlapsed very early. BTW, the model is trained on my own data set. Is there something important??

Answer 5 · 2017-12-25T09:44:03.000Z

Use train_no_tanh.py instead.

Answer 6 · 2017-12-26T09:04:26.000Z

Thx!! It works. I can successfully obtain meaningful images.