https://github.com/lukas-blecher/colorization

Comparing classification and GAN approaches for the colorization problem
https://github.com/lukas-blecher/colorization

Last synced: 10 months ago
JSON representation

Comparing classification and GAN approaches for the colorization problem

• Host: GitHub
• URL: https://github.com/lukas-blecher/colorization
• Owner: lukas-blecher
• Created: 2019-09-14T15:40:10.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2019-10-02T13:31:55.000Z (over 6 years ago)
• Last Synced: 2025-02-08T09:37:27.289Z (11 months ago)
• Language: Python
• Homepage:
• Size: 26.5 MB
• Stars: 4
• Watchers: 4
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Colorization

Final Project in Advanced Machine Learning 2019

In this repo we looked at the colorization problem from two different angles and compare them in our report. 

 * Using a Generative Adversarial Networks

 * Using the classification approach

 

 

 ## Training

To train a model execute a line like the following but replace `train.py` with the corresponing python file (`train_gan.py` or `train_classification.py`)

 `python train.py -n NAME`

 

 ## Results

 

 ![classification results](https://github.com/lukas-blecher/Colorization/blob/master/figures/classification_good.png?raw=true)

 ![gan_results](https://github.com/lukas-blecher/Colorization/blob/master/figures/good-gan-images.png?raw=true)

 

 

 Here are some results where the colorization was successful. 

 Left: Classification, right: GAN  

 The order is from left to right: Grayscale input image, ground truth image, colorized version  

 The images were taken from the [STL-10](https://cs.stanford.edu/~acoates/stl10/) testset.