PyrResNet

For cvpr2018 paper "Intrinsic Image Transformation Via Scale Space Decomposition" Some of the implementations are redundant and will be optimized in the future.

Prerequisites

• Linux(16.04)
• Python 2 or 3
• NVIDIA GPU(TiTan Xp) + CUDA CuDNN

Getting Started

Installation

• Install PyTorch(0.3.0) and torchvision from http://pytorch.org and other dependencies. You can install all the dependencies by

pip install -r requirements.txt

Note: The current software does not update with the newest PyTorch version, some warnings may exist.

• Clone this repo:

git clone https://github.com/liygcheng/PyrResNet.git
cd PyrResNet

• Download Sintel Dataset and MIT Dataset( and additional dataset)

https://drive.google.com/open?id=1gcNSwkDSQCwr8CezgRvL0WOX9wetlFIk

train/test

• Download dataset (e.g. sintel):
• Train a model (e.g. Scene Split):

python PyrResNet_Joint_MPI.py --cuda --niter=1000 

• To view training results and loss plots, change directory to PyrResNet/Results/ and start tensorboard as:

tensorboard --logdir=LossVis --port=10240

open the URL http://localhost:10240 and you will get the visualization results.

Citation

If you use this code for your research, please cite our papers.

@InProceedings{Cheng_2018_CVPR,
author = {Cheng, Lechao and Zhang, Chengyi and Liao, Zicheng},
title = {Intrinsic Image Transformation via Scale Space Decomposition},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2018}
}