pix2pix HD model for colorization task

• original repo: https://github.com/NVIDIA/pix2pixHD
• paper: https://tcwang0509.github.io/pix2pixHD/
• usage: train own datasets for colorization task without prepare two datasets "_A" & "_B"
• main advantage: prepare datasets "datasets/<project_name>/" , with two folders "train_img" and "test_img"
• why convenient: when own dataset has lots of images, two folders gray and color will cost spaces

Getting Started

Installation

• Install PyTorch and dependencies from http://pytorch.org
• and also below required libraries.

pip install -r requirements.txt

• Training:

bash ./scripts/train_color.sh
# need change dataroot dir
# 2 epoch for time estimation

• Test:

bash ./scripts/test_color.sh
# need change dataroot dir
# 2 epoch for time estimation

TODOs

• vgg loss
• fix test process GPU no used.