Mario Klingemann

Ah thanks for the clarification! Right now I am only dealing with a few hundred training examples, so maybe that's why I didn't notice any speed difference. And yes, if...

You could try to put ``` require 'cunn' require 'cudnn' ``` before the torch.load() line.

Yes, same here. Anyone found a patch for this?

Since the error occurs in visualizer.save_current_results one temporary patch is to comment out the part that calls it in train.lua (line 154): ``` if counter % opt.save_display_freq == 0 then...

One other little thing that is probably obvious, but it took me a bit to figure out - if you are running the notebook locally the path to the "./models"...

If you have a GPU with enough memory and have plenty of time you could train a net with a 1024x1024 input size and just center your 640x480 images -...

What I do is a rather naive approach - I calculate the RGB mean value of the input tile and then the RGB mean of the output tile, then I...

The usual way to prevent that is to add CUDA_VISIBLE_DEVICES=0 to your line that calls the .lua training script (meaning CUDA_VISIBLE_DEVICES=[index of GPU starting from 0]) - like that Torch...

I've looked a bit deeper into the outputs of the convolutional layers and whenever the artifact appears there is indeed a single value outside of the typical numeric range which...

I have the suspicion that these artifacts might be caused by the SpatialBatchNormalization when using the default batchSize of 1. Since I am training my models with a bigger batchSize...