mashmawy / convnetcs Goto Github PK

Hi! Recently, I became interested in neural networks due to the discovery of a service, that allows you to automatically paint pictures. Having studied materials a little, I found out, that the network is written on Caffe. Can you help me to rewrite this neural network for using in ConvNetCS? You've managed to port the model VGG16! You can find this network here: https://github.com/richzhang/colorization

Implement L-BFGS-B optimizer.

Hello! I was able to implement two new layers for ConvNetCS. These layers AvgPoolLayer - layer averaging pooling and AvgUnpoolLayer - layer upsampling. Also I was able to implement the fragment algorithm Gatis for styling images - ContentLoss. It uses ConvNetCS and VGG16. I have problems with StyleLoss-style transfer. I suggest You implement the stylization algorithm Gatis. This would allow the training of a neural network for fast styling and styling images under any styles.

Layers.zip

ContentLoss.zip

Hello! Why not implement in C# a quick algorithm to the styling of the images? At the moment there are many implementations, in addition to C#. Here are a few examples:
In JavaScript: https://github.com/reiinakano/fast-style-transfer-deeplearnjs
In TensorFlow: https://github.com/lengstrom/fast-style-transfer
In Torch7: https://github.com/jcjohnson/fast-neural-style
I believe the C# implementation will be a major breakthrough!