MAlberts99
Why Test result has red spot?
Hi, First thanks for your Implement for AdaIn. I want to know why I get the following result when I use the 'Style Transfer Test.ipynb'. I used the pretrained mode of vgg and decoder as read.md.
As you can see, the result image has many red spots and I don't know why, can you explain about it? thanks By the way,I delete following part in processing of network output: "out = torchvision.utils.make_grid(out.clamp(min=-1, max=1), nrow=3, scale_each=True, normalize=True)" And Use: img = toPIL(out[0]) to get the red spotted image And these red parts is the black color in the origin content image. is there some relation?
In my opinion, the reason for these red spots may come from that some pixels contain extreme values, such as exactly 1 or -1.
So you can try to remove the 'clamp' operation in the code, like:
"out = torchvision.utils.make_grid(out, nrow=3, scale_each=True, normalize=True)"
This may influence the performance of the model to some extend, but it can avoid the red spots in my case.
Besides, I also suggest you reading the original code at https://github.com/xunhuang1995/AdaIN-style
