Adjust parameters in generate.py as opposed to train.py about chainer-fast-neuralstyle HOT 2 OPEN

That wowuld be cool, yeah. Have you read the paper ?

Trick : extend the image, apply the filtrer, reduce the image. You'll reduce spatial frequency of the filter (the style image will appear reduced) but you'll loose speed. You can also do the other way around but you'll lose quality.

from chainer-fast-neuralstyle.

I haven't read the paper, do you think that could help me understand ways to accomplish my goal?

That's an interesting idea. Unfortunately I'm already maxing out my gpu's ram with the current image sizes, so I definitely can't run the style on any larger images.

A problem with the output images is that they have too much style applied to the foreground objects and it makes things a bit too unrecognizable. I was thinking a way to reduce "style" while still applying the color components of the style would be to choose very simple style images, i.e. a few vertical colored rectangles next to one another. Do you think that could yield the color without so much style?

from chainer-fast-neuralstyle.