Counting 2,899 Big Data & Machine Learning Frameworks, Toolsets, and Examples...
Suggestion? Feedback? Tweet @stkim1

Automatic Image Colorization

Introduction

This project implements a deep convolutional neural network for automatic colorization, the problem of converting grayscale input images into colored images. The model is based on the ResNet-18 classifier and trained on the MIT Places365 database of landscapes and scenes. It is inspired by previous work from Iizuka et al., Zhang et al., and Larsson et al. on image colorization. Instructions for colorizing images using the provided pre-trained weights or training the network from scratch may be found below. All code is open-source and implementation details are described in the project report.

Feel free to reach out to me (lukemelas) for any questions regarding the model!

Demos

Here are the results of the network on validation images from the Places365 Dataset: Colorization Results

Here, we use the network to colorize a black-and-white Charlie Chaplin film. The model is applied frame-by-frame, with no temporal smoothing applied:

Colorizing Charlie Chaplin

Colorize Images

Instructions coming soon...

Train a Model

Instructions coming soon...