Deep Convolutional Generative Adversarial Nerworks with Caffe Implementation

This project uses DCGAN to implement gray image colorization. The network copys from aleju/colorizer.

Changed Files:

caffe.proto:
	Add the new line:
		  optional bool param_propagate_down = 6;

net.cpp
	Add the following lines:
		if(param_spec->has_param_propagate_down()){
   	 		param_need_backward = param_spec->param_propagate_down();
  		}

Usage:

1. The Installation completely the same as Caffe. Please follow the installation instructions.
   • Make sure you uncomment WITH_PYTHON_LAYER := 1 to support for python layer. And don't forget make pycaffe
   • Add the ~/caffe_GAN/caffe_GAN/gan_example/lib/layers path to $PYTHONPATH
2. Building the dataset:
   • Download Labeled Faces in the Wild and extract it somewhere
   • In gan_example/ run mkdir out_unaug_64x64 and run python lib/utils/generate_dataset.py --path="lfw", where lfw is the path to your LFW dataset
   • Generate the train.txt file:
   • In out_unaug_64x64/ run ls -1 > ../train.txt
3. Train
   • gan_example/ run ./train.sh begin training.
4. The generate images will be putted in output directory.