search

jalola / improved-wgan-pytorch

Improved WGAN in Pytorch
MIT License
439 stars 68 forks source link

Increasing dimensionality #27

Closed alexandreberkovic closed 2 years ago

alexandreberkovic commented 2 years ago

Hi, I was taking a look at the model and saw several hard-coded 64s but those both represent batch_size and model dimensionality, which ones should be changed in order to increase dimensionality of the GAN ?

jalola commented 2 years ago

Hi, You can try python train.py --help to see options it supports, there is an option dim which is 64 by default. I remember I tried 128, it gives me Out Of Memory for me and when I tried 96, it has some wrong size error. In case it has error because of the size, you need to modify the ResidualBlock in both GoodGenerator and GoodDiscriminator

alexandreberkovic commented 2 years ago

That makes sense, would you add a ResidualBlock layer to both Generator and Discriminator to make it work then ?

jalola commented 2 years ago

This is already supported as I said, just use --dim SIZE so you can train with bigger dimensionality. I just updated some requirement to support higher Pytorch version.