Does the discriminator really work？

agrimgupta92 / sgan

Code for "Social GAN: Socially Acceptable Trajectories with Generative Adversarial Networks", Gupta et al, CVPR 2018

MIT License

826 stars 262 forks source link

Does the discriminator really work？ #83

Open xxAna opened 4 years ago

xxAna commented 4 years ago

Thank you for your sharing!That's a good job! And when I run the code,I found that the loss of discriminator seems don't converge, and the loss of the gennerator is mainly depend on l2 loss. Is that normal? Waiting for your reply. I am a new beginner. Please correct me if I am wrong. Thank you again and best wishes!

zephyr-fun commented 4 years ago

Thanks for your sharing! I also found this problem when I was running the code, and it feels like the discriminator is not converging much, is this normal? If this is normal, then why does the discriminator look like it's not trained?

SecretMG commented 4 years ago

Excuse me, could you tell me how to draw training curves like yours? Is it tensorboardX?

xxAna commented 4 years ago

@SecretMG yes

cxc1219 commented 4 years ago

@SecretMG yes

Have you attached other codes?