How can I make the latents in data_latents and latents_ in edits_sg?

[naeun0620]

Hi, I have a question. It would be really helpful if you answer me.

I am wondering how to make the *.npy s in data_latents and latents_ folders.

For example, If I want to make latents_/beard_glass.npy and data_latents/men.npy for edit_type 'beard', what should I save?

Is the normalized B in https://github.com/RameenAbdal/CLIP2StyleGAN/blob/main/optimize_dir.py#L85 can be latents_/beard_glass.npy?

Also, how can I make data_latents/men.npy? Can I just get all the men images and extract features like the code below?

   image_data = natsorted(glob.glob(args.path))
   with torch.no_grad():
      for i in range(len(image_data)):
         if i == 0:
            i_x = preprocess(Image.open(image_data[i])).unsqueeze(0).to(device)
            image_features_pred = model.encode_image(i_x)
         else:
            i_x = preprocess(Image.open(image_data[i])).unsqueeze(0).to(device)
            image_features_pred = torch.cat((image_features_pred, model.encode_image(i_x)),dim = 0)

Thanks,