def infer_batch(feed_im: torch.Tensor):
feed_im = feed_im.to(device)
out_im = model.forward(feed_im)
draw_im = (denormalize(
out_im.permute((0, 2, 3, 1)).detach().to('cpu').numpy()) *
255).astype('uint8')
return draw_im
def color_loss(self, con, fake):
con = rgb2yuv(denormalize(con))
fake = rgb2yuv(denormalize(fake))
return (self.l1_loss(con[..., 0], fake[..., 0]) +
self.huber_loss(con[..., 1], fake[..., 1]) +
self.huber_loss(con[..., 2], fake[..., 2]))
def _process(self, x):
# NOTE 图像范围为[-1~1]，先denormalize到0-1再归一化
return self.vgg_normalize(denormalize(x))
def _process(self, x):
# NOTE 图像范围为[-1~1]，先denormalize到0-1再归一化
rgb = denormalize(x) * 255 # to 255
bgr = rgb[:, [2, 1, 0], :, :] # rgb to bgr
return self.vgg_normalize(bgr) # vgg norm