Developing systems which enable us to colorize diversely
SerialLain3170 opened this issue · comments
I have implemented architectures which generate colorization of single style. However, considering MUNIT, DRIT and so on, I'm wondering if I can develop systems which output diverse colorization. As far as I know, it seems that no papers about diverse colorization have existed. If anybody have ideas or know papers, please let me know.
I try to implement some methods which are related to disentanglement.
I have tried diverse colorization without hint using BicycleGAN + Mode seeking regularizer in the directory.
However, I would like to develop diverse colorization with hints. In other words, I want to exploit the method for producing illustrations that have diverse drawing styles with the same color. I could not do this with using BicycleGAN, so I need to seek for another method.
well, the best way I know about for this task is using stylegan inversion (if you only want to colorize a constrained domain,like anime head), check this paper: Encoding in Style: a StyleGAN Encoder for Image-to-Image Translation
Thanks, I'll check it! Like thie paper, I have noticed many papers incoporating StyleGAN in I2I tranlation. StyleGAN has a great versatility...
Hi Colorization Transformer can generate diverse colorization, maybe you can see it
Thanks! I have found the idea of Colorization Transformer interesting.