sandeep-krishnamurthy commented on issue #12582: In dcgan.py, the parameters of netD are updated as training G? URL: https://github.com/apache/incubator-mxnet/issues/12582#issuecomment-436434362 Hello @thbupt I verified that as expected, with trainerG.step(), the Discriminator network (netD) is not updated. Below is the code snippet I used in the training loop under update G network. ```python with autograd.record(): fake = netG(latent_z) output = netD(fake).reshape((-1, 1)) errG = loss(output, real_label) errG.backward() print("Before ") print(netD(fake)) trainerG.step(batch.data[0].shape[0]) print("After ") print(netD(fake)) ``` However, please note that you are generating a new fake images here. You are not using fake images generated in Update D network section. So, if you verify netD(fake) before this block, you are actually using different input fake images so the output is different. Resolving for now. Please reopen if issue still persists or you have any further queries.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
