Valentin

@LouiValley, this one? https://github.com/BVLC/caffe

@LouiValley, I guess, you mean something like this, but for "neural style": https://github.com/msracver/Deep-Image-Analogy There are some neural projects reviews, and as far as I can tell, no one tried to...

@fastlater, [here](https://handong1587.github.io/deep_learning/2015/10/09/fun-with-deep-learning.html#neural-art) are at least 2 projects of normal style transfer in TensorFlow: [by @anishathalye](https://github.com/anishathalye/neural-style) and [by @ckmarkoh](https://github.com/ckmarkoh/neuralart_tensorflow). As for fast style transfer, the style and content weights are saved...

@fastlater, come to think of it, _maybe_ there is a way to make fast style transfer with variable strength: https://github.com/Heumi/Fast_Multi_Style_Transfer-tensorflow You can make multi-style model with two styles: - 1st...

**ProGamerGov**, just a little suggestion: since GPU handling is already implemented in ["function setup_gpu(params)" (_line 324_)](https://github.com/ProGamerGov/neural-style/blob/6814479c8ebcc11498b7c123ee2ba7ef9f0fe09f/neural_style.lua#L324), maybe it's possible to use that function instead of new "`set_datatype(data, gpu)`"? It could...

@ProGamerGov, yes. And to delete function `set_datatype(data, gpu)` at [line 611](https://github.com/ProGamerGov/neural-style/blob/6814479c8ebcc11498b7c123ee2ba7ef9f0fe09f/neural_style.lua#L611), as it will not be needed anymore.

@htoyryla, reflective padding probably takes pixels starting from 1 pixel distance: [ x-2, x-1, x ] _[ x-1, x-2 ]_. And replication duplicates the edge: [ x-2, x-1, x ]...

After making more tests with different models, I was wrong: the noise is not added by padding. It's a quality of some models: vgg19 from [crowsonkb's repository](https://github.com/crowsonkb/style_transfer) makes clean images...

@ProGamerGov, by the way, the code can probably be reimplemented in Torch and maybe even [will not be too different](https://github.com/torch/torch7/wiki/Torch-for-Numpy-users): Python | Torch -- | -- [ndarray.shape](https://docs.scipy.org/doc/numpy/reference/generated/numpy.ndarray.shape.html) | [size](https://github.com/torch/torch7/blob/master/doc/tensor.md#torch.Tensor.size) [numpy.mean](https://docs.scipy.org/doc/numpy/reference/generated/numpy.mean.html)...

Speaking of "vgg_normalised.caffemodel": together with much higher style weight, it probably _requires_ "normalize_gradients". In above examples with this model, the content weight was 100, style weight 300000, and gradients were...