======================================================================== INTRODUCTION Fully Automatic Image Colorization Code Created by Zezhou Cheng (chengzezhou@sjtu.edu.cn) This MATLAB code implements the colorization algorithm presented in the following paper: Zezhou Cheng, Qingxiong Yang, and Bin Sheng. Image Colorization Using Neural Networks. Submitted to IEEE Transactions on Image Processing 2016. The preliminary version of this work is described in: Zezhou Cheng, Qingxiong Yang, and Bin Sheng. Deep Colorization. Proceedings of the IEEE International Conference on Computer Vision. 2015: 415-423. ======================================================================== INSTALLATION This code was tested on MATLAB 2015b under UBUNTU 14.04.4 LTS If you want to colorize your own images, you need to install the following package: Caffe : https://github.com/longjon/caffe/tree/future OR https://github.com/BVLC/caffe/ Please refer to http://caffe.berkeleyvision.org/installation.html ======================================================================== USAGE (1) For the experiment on your own test images : Single image colorization(default name of image 'test.jpg') : --- run demo_single.m Batch colorization (default path of images 'testset'): --- run demo_batch.m (2) For a quick experiment on the bundled examples without the requirement of installing CAFFE: --- run run_examples.m ======================================================================== LIBRARY The following libraries are included in this code : (1) DAISY Descriptor http://cvlab.epfl.ch/software/daisy (2) GIST Descriptor http://people.csail.mit.edu/torralba/code/spatialenvelope/ (3) Fully Convolutional Networks For semantic Segmantation https://github.com/shelhamer/fcn.berkeleyvision.org Note that we retrained this model on grayscale version of images from SIFT Flow Database. Our trained model is placed in 'scene_parsing_model' folder, named as 'semantic_segmentation.caffemodel' (4) Domain Transform RF filter http://inf.ufrgs.br/~eslgastal/DomainTransform/