ProGamerGov

Are these the same artifacts?   Because spatial decorrelation (FFT parameterization) and color decorrelation seem to make them worse, and transform robustness just ends up creating outs that aren't...

@genekogan I have implemented spatial decorrelation, color decorrelation, and transform robustness into neural-style-pt here to see if they could help resolve the artifacts, but it appears to have them so...

So, the Mona Lisa image uses the sRGB IEC61966-2.1 color profile and when PIL loads it and converts it to an RGB image there's a slight change in the colors....

@genekogan Since your artifacts don't seem to resemble the checkerboard artifacts from conv layers, maybe the issue comes from the pooling layers? Figure 3 from this research paper: https://arxiv.org/abs/1511.06394 seems...

So, I can implement my own version of MaxPool2d in one of two ways: ``` class MaxPool2d(torch.nn.MaxPool2d): def forward(self, x): x = x.unfold(2, self.kernel_size, self.stride).unfold(3, self.kernel_size, self.stride) x = x.contiguous().view(x.size()[:4]...

@gateway A specific seed value would let you recreate the same stylized video. Specific seeds aren't required for stylizing video. I haven't seen anyone using neural-style-pt with proper optical flow...

@gateway PyTorch is light years ahead of other frameworks like TensorFlow in terms of how easy it is to do things. Python is also a pretty easy to learn language,...

@yunlm Try using the following parameters with multiscale generation: ``` -init image -tv_weight 0 -learning_rate 2 -init image -content_weight 100 -style_weight 6000 ```

@strohrbaugh You should start off at a lower image size and then scale it up with multiple steps of increasing image size (multiscale rendering).

@spot92 I need more information to know if it's a neural-style-pt issue or a PyTorch/CUDA issue. If it's a PyTorch issue with the latest update, then trying to to simplify...