The training codes of DCGAN to train CIFAR10

[LiuzhuForFun]

Thanks for this excellent work！ But I met one problem to train DCgan on CIFAR10. Could you please to give me some details of this configs of training?

[kwotsin]

Hi @LiuzhuForFun, thanks for your feedback! For training DCGAN, I follow the original configurations for the reported score [1], and specifically from Table 10 and section 2.6 of [1]:

• Learning rate: 2e-4 for D and G.
• (beta_1, beta_2) = (0.5, 0.999) for Adam optimizer
• 5 discriminator updates per generate update
• Train for 200k steps, no learning rate scheduling.
• Batch size 64

Note that the model uses dcgan_cifar.py and the non-saturating loss, following the original configuration

Hope it helps! :D

[1] https://arxiv.org/pdf/1807.04720.pdf