- Creation: 01/30/2020
- Update: 02/03/2020
In The 16th ACM SIGGRAPH European Conference on Visual Media Production. Yliess Hati, Grégor Jouet, Francis Rousseaux and Clément Duhart
The lack of information provided by line arts makes user guided-colorization a challenging task for computer vision. Recent contri-butions from the deep learning community based on GenerativeAdversarial Network (GAN) have shown incredible results com-pared to previous techniques. These methods employ user inputcolor hints as a way to condition the network. The current state ofthe art has shown the ability to generalize and generate realisticand precise colorization by introducing a custom dataset and a newmodel with its training pipeline. Nevertheless, their approach relieson randomly sampled pixels as color hints for training. Thus, inthis contribution, we introduce a stroke simulation based approachfor hint generation, making the model more robust to messy inputs.We also propose a new cleaner dataset, and explore the use of adouble generator GAN to improve visual fidelity.