educational-challenge-image-colorization

Repository for the Image Colorization educational challenge

Design Doc

Install

Prerequisites

Make

In order to run make commands, install:

sudo apt-get install --reinstall build-essential

Python Environemnt

conda env create -f environment.yml
conda activate ecic

Download dataset

To download and assemble the dataset, use:

make all

This will create a directory ./data/train. Note that the images require lots of storage, so make sure you have that available first. The dataset is a combination of Places365 and OpenImages by tencent.

Generate gray scale images

For convenience, we also added a script to convert all the images to black and white. In order to generate black and white images, just run make generate-black-white.

Alternatively, we strongly encourage you to try out different methods to gray scale the images, and hence get better models that perform well on different type of gray-scale modification.

Here's the method we are using (it uses Pillow)

from PIL import Image

def grayscale_image(input_path, output_path):
  img = Image.open(input_path).convert('L')
  img.save(output_path)