The pre-trained sytleGAN2 discriminator can not discriminate fake and real images

[chobao]

Hi, I want to use the pre-trained styleGAN2 discriminator as the supervision for training another image generation network. However, the pre-trained discriminator always outputs negative scores ( denotes fake images) whenever real or fake images are fed. Is it normal in GAN? Or is there a mistake in my usage? Here are my codes.

def fit_to_stylegan_range(x):
    return  2 * x - 1  # Scale from range (0, 1) to range (-1, 1)

for path in files: 
        name = os.path.basename(path)
        image =  torch.from_numpy(np.asarray(Image.open(os.path.join(path)))) / 255.0
        image = image.unsqueeze(0).permute(0, 3, 1, 2)
        image = torch.nn.functional.interpolate(image, size=(512,512), mode="bicubic")
        min_i = image.min()
        max_i = image.max()
        image = (image - min_i) / (max_i - min_i)
        print("image min:{}, max:{} ".format(image.min().item(), image.max().item()))

        image = fit_to_stylegan_range(image) 
        d_fake = discriminator(image)
        prob_fake = F.softplus(d_fake)
        print("{} discriminator: {}({})".format(name, prob_fake.item(), d_fake.item()))

[gonzalosilva99]

@chobao Hi! I know your question was long time ago, but I am facing the same issue. Were you able to solve it?