Deep Koalarization
Image Colorization using CNNs and Inception-ResNet-v2 (2017)

Abstract

We review some of the most recent approaches to colorize gray-scale images using deep learning methods. Inspired by these, we propose a model which combines a deep Convolutional Neural Network trained from scratch with high-level features extracted from Inception-ResNet-v2 pre-trained model. Thanks to its fully convolutional architecture, our encoder-decoder model can process images of any size and aspect ratio. Other than presenting the training results, we assess the “public acceptance” of the generated images by means of a user study. Finally, we present a carousel of applications on different types of images, such as historical photographs.

Authors

Federico Baldassarre

@baldassarrefe

Lucas Rodés-Guirao

@lucasrodes

Diego González Morín

@diegomorin8

Code

A TensorFlow/Keras implementation of Deep-Koalarization is available on Github.

Watch Star Fork

Paper

 
@article{deepkoal2017,
  author          = {Federico Baldassarre, Diego Gonzalez-Morin, Lucas Rodes-Guirao},
  title           = {Deep-Koalarization: Image Colorization using CNNs and Inception-ResNet-v2},
  journal         = {ArXiv:1712.03400},
  url             = {https://arxiv.org/abs/1712.03400},
  year            = 2017,
  month           = dec
}

arXiv e-print

Acknowledgement

We would like to thank Prof. Josephine Sullivan for supervising our work. Furthermore, we also want to point out that our network was trained and tested using the Tegner nodes of the PDC Center for High-Performance Computing at the KTH Royal Institute of Technology, leveraging the NVIDIA® CUDA® Toolkit and the NVIDIA® Tesla® K80 Accelerator GPU to speed up the computations.

home