Image colorization is the task of assigning colors to a grayscale image such that the resulting colored image is visually pleasing and semantically meaningful. We implemented the model proposed by “Deep Koalarization: Image Colorization using CNNs and Inception-ResNet-v2”[1] paper and explored a few modifications to the architecture. We also explored transfer learning using different pre-trained models. Our results show that the model is able to successfully color high-level image components such as the sky, the sea, or forests. However, the performance in coloring small details is still to be improved. We believe that further improvements can be made by training the network over a larger training dataset.
For detailed information, please refer to the report.
[1] Baldassarre Fe, Diego Gonz´alez Mor´ın, and Lucas Rod´es-Guirao. ”Deep Koalarization: Image Colorization using CNNs and Inception-Resnet-v2.” arXiv preprint arXiv:1712.03400 (2017).