Pytorch-TextureGAN icon indicating copy to clipboard operation
Pytorch-TextureGAN copied to clipboard
verified publisher icon janesjanes

Parameter Issue

Open Pengxiao-Wang opened this issue 7 years ago • 3 comments

Hi, Thank you for this research — I was confused with the parameter differences between the original paper and what you ilustrated in README (e.g. the weight of each loss, lr ...). Above all, was the parameter "local_texture_size" supposed to be '-1' in the first training state, which you took '50' in both states?

Pengxiao-Wang avatar Nov 19 '18 11:11 Pengxiao-Wang

Hi, I got the same problem, and still failed to train a useful model which can provide the similar output as that given by the official model. Do you guys have any suggestions or hints on parameters and required epochs?

Thanks,

HideUnderBush avatar Jan 14 '19 12:01 HideUnderBush

@Pengxiao-Wang @HideUnderBush Have you got the solution for "style_weight" and other parameters, what should be their values?

chandrakant-sonawane avatar Feb 11 '19 09:02 chandrakant-sonawane

Hi, I got the same problem, could you share your parameters for me at pre-train and fine-tuning? @Pengxiao-Wang @HideUnderBush @chandrakant-sonawane

kmaeii avatar May 07 '19 09:05 kmaeii