how to convert sketch into fixed_z (.pth file) when running generate.py

[jingnian-yxq]

I tried pix2latent to get fixed_z (.pth file), but met several problems:

1. after running invert_stylegan2_*.py, a .npy file is saved for variables, but how can I get a .pth file?
2. the repo for stylegan2 is for size 384x512, while my sketch is 256x256, thus it reports many errors for wrong dims. I don't know how to correct it.
   Thanks a lot and looking forward to responding.

[PeterWang512]

I think pix2latent is more suitable when your input is an RGB image. The idea is the get the latent z that captures the texture, color, and background of the image, and then edit the shape by changing the model weights. Hope this helps!

[jingnian-yxq]

Thanks for your reply! I am wondering how can I generate an image from a given sketch input by running generate.py, since the samples are generated without user sketch input. I viewed some issues here and found pix2latent is mentioned for getting fixed_z for the input of generate.py, but I still don't know how to get it, for the file type as well as the image size :(

[PeterWang512]

Our work focuses on creating a model based on a sketch, so you'll need to train a model with the sketch. Instruction is specified in this section: https://github.com/PeterWang512/GANSketching#model-training

After that, you will have models that generate infinite images with the same shape and pose as the sketch. Hope this helps!

[jingnian-yxq]

I'm confused. I have completed the model training with my sketch. Then after the model is completed, how can I generate images with another sketch(not training sketches) by applying the trained model? In fact, this is what I want to do.