yuxinn-j / scenimefy Goto Github PK

Thanks for your work,

1. Will you post more models
2. The test portrait effect is very poor, why?

Hi! I wonder how long does it take for one group of experiment? Thanks!

Hello, I have a question: why are real_A and real_B spliced ​​together when training unpaired images, but there is no such step when training paired images?
"""
Run forward pass; called by both functions <optimize_parameters> and ."""
self.real = torch.cat((self.real_A, self.real_B),
dim=0) if self.opt.dce_idt and self.opt.isTrain else self.real_A
if self.opt.flip_equivariance:
self.flipped_for_equivariance = self.opt.isTrain and (np.random.random() < 0.5)
if self.flipped_for_equivariance:
self.real = torch.flip(self.real, [3])

    # forward unpaired data
    self.fake = self.netG(self.real)
    self.fake_B = self.fake[:self.real_A.size(0)]
    if self.opt.dce_idt:
        self.idt_B = self.fake[self.real_A.size(0):]

    # forward paired data
    self.fake_p = self.netG(self.real_A_p)
    self.fake_B_p = self.fake_p[:self.real_A_p.size(0)]

Hi, thank you for your greate work! But I don't find where the StylePatchNCE loss, can you help me? I'm looking forward your repying.

Hello 👋 very interesting work! Do you have any plans to release StyleGAN finetuning scripts?

From the project web, those demo videos look very stable and amazing. How to process a video? Just extract a video to frames and process frames one by one? Or there is other stability tech？

Where can I get your test dataset you mentioned in your paper?

Could you please tell me how to train your model for this code?

As you say that the high-quality anime scene dataset comprises 5,958 images, but I follow the Scenimefy/Anime_dataset/README.md, obtaining 5151 cartoon images. If I were to manually filter the dataset again, the number of images used for training would be even smaller. Why did you get 5958 images.

In the README there are the instructions on how to train a custom model. It says to run train.py, but this file is missing.