Instance-aware image colorization by Jheng-Wei Su, Hung-Kuo Chu and Jia-Bin Huang proposes a brilliant idea where a model colorizes a black and white image while being aware of the specific objects in an image along with the entire image.
We (@ayush_thakur and I) have come up with a minimal reproduction of the paper in TensorFlow.
In the repository, we have tried encapsulating all the main features of the training process as suggested in the paper. The repository consists of jupyter notebook only, this was done to help readers with reading the code better and also execute and experiment on it.
What we have covered:
- The three-stage training process
- Ablating on the rgb and the lab color spaces
You can also check this Weights & Biases report for a quick paper summary.
Report: Weights & Biases
Official Colab Notebook: Google Colab
Paper: https://arxiv.org/abs/2005.10825