machine-learning-for-image-colorization's Introduction

Image Colorization

To clone the whole project, use git clone --recursive:
git clone --recursive https://github.com/Lyken17/Image-Colorization

Approach

Gray Output (To test baseline error)
HOG + SVR (HOG_SVR)
Residual Encoder Network (ResidualEncoderNet)
Multi-Level Feature Network (MLNet)
Stylization Network (StylizationNet)

Approachs	Reference	Model Size	Quality
Multi-Level Feature (this one)	siggraph16	332M	High
Stylization Network	[ECCV]	17M	Medium ~ High (best on outdoor scences)
Residual Encoder Network	\	98M	Medium

For Colorizing Images or Training Models

Please refer to the sub-directory corresponding to the specific approach you want to utilize.

Pre-trained Models

Please download them in RELEASE
Technical details

Please refer to our report

Poster

License

See sub-directory. If sub-directory doesn't contain a LICENSE file, then MIT License applies.

machine-learning-for-image-colorization's People

Contributors

Stargazers

Watchers

machine-learning-for-image-colorization's Issues

Siggraph method experiment

Now this experiment is running on GPU01, with microsoft coco dataset.

iteration 4000
Running on validation set ... 	
val loss = 0.018003

Seems it has a better colorization ability.

Training and testing data

Please use the following data for training and testing:

Training: http://msvocds.blob.core.windows.net/coco2014/train2014.zip
Testing: http://msvocds.blob.core.windows.net/coco2014/val2014.zip

Testing Error: (1/3)sqrt((1/n)(1/(256256))sum((y-y0)^2+(u-u0)^2+(v-v0)^2)).
[Average yuv RMS error for each pixel each channel for each testing image.]

upload the training and testing dataset

To compare different methods, we must use the same training and testing dataset. As discussed, testing dataset is Microsoft coco. Should we download some images elsewhere as testing data for evaluation purpose?

I can create a release and collect & upload all training/testing dataset if you haven't done it @Lyken17

Project thesis

style transfer

https://github.com/szagoruyko/torch-opencv-demos/tree/master/texture_nets

age prediction

http://torch.ch/blog/2016/06/01/deep-fun-with-opencv.html

face detection

http://torch.ch/blog/2015/11/13/gan.html

lyken17 / machine-learning-for-image-colorization Goto Github PK