cyclomon / clipstyler Goto Github PK

Hello authors,

What hyperparameters/settings did you used for producing the StyleGAN-NADA? Im having very different results.

Thank you for your atttention.

Hi,
I got the following error while testing the fast style transfer model,

Traceback (most recent call last):
  File "test_fast.py", line 76, in <module>
    network = net.Net(vgg, decoder)
NameError: name 'net' is not define

Seems that net.Net should be fast_stylenet.Net in

There seems to be another typo that --content_path should be changed to --content_dir in README for train_fast.py

https://github.com/cyclomon/CLIPstyler/blob/main/README.md#fast-style-transfer

First I must say, this tool is pure magic! Thanks for developing it.

I see that in the research paper, there is a demonstration image which gets style-transferred to its original format, which was a non-square, it was a image of a field and skies. I also never got any image output over the resolution of 512, which my 24GB card would surely handle.

I tried settings the --crop_size, --num_crop and --image_size options, but it never did anything. I'm no programmer (and still have a github lol, I want to learn rust eventually) and I looked through the various .py files, but couldn't find anything obvious to change it.

Is there not any way to change this in the released model, or did I just miss something?

Hi, I recently saw your work intrigued, but trying to run your training code got the following problem, can you give me a hint?