God said, "let there be color"; he willed it, and at once the Deep Neural Network brings the color!
Here is an intuitive demo.
This project is based on the siggraph16 paper. Please refer to this paper for the details of the model.
Our implementation is slightly different from the one proposed in the aforementioned paper. The experiments show that our implementation is much more efficient and requires significantly less graphic memory in the training phase compared to the implementation proposed by the paper.
More colorization methods are available here Machine-Learning-for-Colorization. Pick one that fits your photo best.
-
Hardware requirements
If you want to train a model on your own dataset, a powerful GPU is required (at least 6GB graphic memory). If you just want to play with it, a CPU would be enough and you should download our pre-trained models.
-
Install the framework
The implementation of this project is on the Torch7, a popular deep learning framework. Follow the install guide to install this framework on your computer.
-
Setup dependencies
luarocks install torch luarocks install nn luarocks install nngraph luarocks install image luarocks install lua-cjson luarocks install hdf5 #GPU acceleration luarocks install cutorch luarocks install cunn luarocks install cudnn -
Download pre-trained model (if you don't want to train your own model)
wget https://github.com/Lyken17/Colorize-Your-World/releases/download/1.0/pre_trained.t7 -
Colorize your favorite images
th colorzie.lua -input_image 'your image'
The pre-trained model we provide is carefully finetuned and should work well for general purposes. But if you want to train your own model anyway, you can use the following command (with the default training parameters):
th train.lua -h5_file 'your training database here'
You can specify many training parameters. Please refer to train.lua for a full list of parameters.