hujan01 / layoutgan Goto Github PK

Hello, I am trying to implement LayoutGAN Demo 2 from your repo for RICO dataset. I have used the semantic_annotations.zip folder and converted each json file into a 9x9 matrix(bounding box coordinates, class probability one hot encoded vector, total size = 4+5=9) as the input to be given to LayoutGAN. I couldn't understand the changes i should make in your demo2 main.py -> points_to_img() function. Can you please tell me what all changes should be made in your current MNIST code?

I am not on GitHub, I created an account to just contact you here because I couldn't find your name or email address anywhere. I have some more doubts related to LayoutGAN and would appreciate it if you can tell me your email id over here so I can send a mail. Or, pleases mail me at [email protected] or message me on linkedin, so I can contact you.

Thanks & Regards

hi, jianh, thanks for your work. Your code is really readable. But after training 200 epochs with your code, I didn't get orderly results at all. What discriminator was used to train it? Which code did your performance get from?
demo2, 200epoch, realation discriminator

demo2, 200epoch, wireframe rendering discriminator

demo3, 20000 step, realation discriminator

demo4, 200epoch, wireframe rendering discriminator

training environment:
python==3.6.3
pytorch==1.0.1.post2
GPU==tesla m40 24G
cuda==9.0

Thank for your efforts, I ran the Demo2 for 76 epoches and I stopped the codes. I dont know how to continue training starting from 76 epoches. Thanks for your reply !