DeepColor

GitHub link

Medium article

Project for colorizing grayscale pictures and comparing 3 different neural networks:

• Deep Koalarization: Encoder-Decoder + Inception Resnet fusion classifier
• Unet with Resnet50 backbone
• Pix2Pix GAN

Versions

• keras (2.8.0)
• matplotlib (3.5.2)
• numpy (1.23.4)
• opencv (4.6.0)
• tensorboard (2.8.0)
• tensorflow (2.8.2)
• tqdm (4.64.1)

How to use

usage: main.py [-h] [-d] [-t TRAIN] [-m MODEL] [-c COLORIZE] [-v VIDEO]

optional arguments:
  -h, --help            show this help message and exit
  -d, --debug           set log level to debug
  -t TRAIN, --train TRAIN
                        Train a model. Choose between 'fusion', 'unet', 'gan'.
  -c COLORIZE, --colorize COLORIZE
                        Colorize a picture at the given path
  -v VIDEO, --video VIDEO
                        Colorize a video at the given path

Some results

Colorizing pictures

Colorizing video

Architectures diagrams

Deep Koalarization

Unet

Pix2Pix