search

richzhang / colorization

Automatic colorization using deep neural networks. "Colorful Image Colorization." In ECCV, 2016.
http://richzhang.github.io/colorization/
BSD 2-Clause "Simplified" License
3.35k stars 928 forks source link

Most colorizations (with people) are brownish #37

Open panuganti opened 7 years ago

panuganti commented 7 years ago

Quite a lot of grayscale pictures (taken from black & white movies) are colorized into brownish color. How that can be improved ?

microic commented 6 years ago

I also meet this problem, “brownish” means no overall feature is learned see https://github.com/microic/niy/tree/master/examples/colorizing_photos

richzhang commented 6 years ago

The system is trained with modern day photographs. Black&white stills from movies may look very different than the training set. This is known as a "domain shift." A possible fix would be to fine-tune the system using stills from older color movies.

panuganti commented 6 years ago

@richzhang Do you have a keras (or TFLearn) implementation of your amazing project ? I'd like to try training the system using the stills from those old black & white movies.

richzhang commented 6 years ago

I do not, but there may be other implementations out there!