Code for executing the Single-Image Intrinsic Decomposition algorithm presented in the paper:

Unsupervised Deep Single-Image Intrinsic Decomposition using Illumination-Varying Image Sequences

by Louis Lettry, Kenneth Vanhoey and Luc Van Gool

in Computer Graphics Forum, vol. 37, no. 10 (Proceedings of Pacific Graphics 2018).

Paper available on kvanhoey.eu and ArXiv.

Please cite our paper if you use, compare to, or get inspired by this code and/or work.

@article{LVvG18,
    author={Lettry, Louis and Vanhoey, Kenneth and {Van Gool}, Luc},
    title="{Unsupervised Deep Single-Image Intrinsic Decomposition using Illumination-Varying Image Sequences}",
    booktitle={2018 IEEE Winter Conference on Applications of Computer Vision (WACV)},
    year = "2018",
    month = "October",
    journal = "Computer Graphics Forum (Proceedings of Pacific Graphics)",
    volume = "37",
    number = "10"
}

This repository contains two python scripts. The main entry point is main.py. It

1. loads the pre-trained neural network, and
2. batch-executes it on a list of files.

An example file is given in the folder input/ Global parameters at the top of main.py allow to edit location of input and result folders. The folder model/ contains the pre-trained CNN definition and weights.

1. Clone this repository

git clone [email protected]:lettryl/UnsupervisedIntrinsicDecomposition.git

2. Enter the repository folder

cd UnsupervisedIntrinsicDecomposition/

3. (Optional) place the files you want to process in the folder input/
4. Run the application

python main.py

Tested on Linux Ubuntu 18.04. Should probably work on other systems with minor effort.

Depends on:

• Python (developed on 2.7, compatible with Python3)
• Tensorflow (developed on v1.2.0)
• Python libraries: NumPy, PIL.