Awesome

Tensorflow implementation of Deshpande et al. "Learning Diverse Image Colorization"

The code is tested for Tensorflow-v1.0.1 and python-2.7. The code additionally needs numpy, scipy, scikit-learn and caffe-r1.0 (caffe only for Zhang et al. colorization network).

Fetch data by

bash get_data.sh

Fetch Zhang et al. colorization network for MDN features by

bash get_zhang_colorization.sh

Execute run_lfw.sh to first train vae+mdn and then, generate results for LFW

bash run_lfw.sh

Execute run_demo.sh to get diverse colorization for any image, the model is trained on imagenet

bash run_demo.sh

If you use this code, please cite

@inproceedings{DeshpandeLDColor17,                                                                  
  author = {Aditya Deshpande, Jiajun Lu, Mao-Chuang Yeh, Min Jin Chong and David Forsyth},          
  title = {Learning Diverse Image Colorization},                                                    
  booktitle={Computer Vision and Pattern Recognition},                                              
  url={https://arxiv.org/abs/1612.01958},                                                           
  year={2017}                                                                                       
} 

Some examples of diverse colorizations on LFW, LSUN Church and ImageNet-Val dataset

Some examples of diverse colorizations for images in the wild, model is trained on imagenet