A Neural Network For Automatic Image Colorization
This project is a PyTorch version of the ColorNet issued on SIGGRAPH 2016. Please check out the original website for the details.
-
Net model
-
DataSet MIT Places205
Hint: For there are grayscale images in the dataset, I write a script to remove these images
- Development Environment
Python 3.5.1
CUDA 8.0
I just train this model for 3 epochs while 11 epochs in the paper， so I think it will work better if train it more.
- Good results
- Bad results
For this network is trained by landscape image database, it's work well for scenery pictures. So if you use this network to color images of other types, maybe you can't get a satisfying output.
You can download the model from https://drive.google.com/file/d/0B6WuMuYfgb4XblE4c3N2RUJQcFU/view?usp=sharing
Implement with PyTorch1.0