Automatic image colorization, which means colors greylevel images, has been a hot field due to its valuable applications like old-paragraph colorization. Traditionally, this work needs people to set color value to each pixel. Hopefully, deep learning can provide tremendous convenience to image colorization. For instance, Iizuka (Iizuka, Simo-Serra, and Ishikawa 2016) proposed a fully automatic image colorization based on Convolution Neural Networks (CNN) in an end-to-end fashion. Isola et al. (Isola et al. 2017) introduced Generative Adversarial Networks (GAN) to image translation, including image colorization. Usually, a deeplearning model predicts a color value to each pixel of the grayscale image.
This work is to implement the above two approachs and make comparsion.
DL_Colorization_GAN_v1_0.ipynb with full output is large to render in Github, so the uploaded version is with part of output.