iTMNet

This is an implementation for iTM-Net: Deep Inverse Tone Mapping Using Novel Loss Function Considering Tone Mapping Operator.

Project page, DOI

When you use this implementation for your research work, please cite the paper.

The following is the bibtex entry.

@article{kinoshita2019itmnet,
author = {Kinoshita, Yuma and Kiya, Hitoshi},
doi = {10.1109/ACCESS.2019.2919296},
issn = {2169-3536},
journal = {IEEE Access},
volume = {7},
number = {1},
pages = {73555--73563},
title = {{iTM-Net: Deep Inverse Tone Mapping Using Novel Loss Function Considering Tone Mapping Operator}},
url = {https://ieeexplore.ieee.org/document/8723346/},
month = {May},
year = {2019}
}

Requirements

• Python 3.9 or later

• Pytorch 1.8 or later

• hdrpy 0.1.0 or later (in the external directory)

  Repo: https://github.com/popura/hdrpy

• deepy 0.5.0 or later (in the external directory)

For other requirements, see pyproject.toml

Getting started

1. Clone this repository

   git clone https://github.com/popura/itmnet-pytorch.git
   cd itmnet-pytorch
   

2. Install requirements.

   If you use poetry as a package manager, it is done by

   poetry install
   

3. Prepare a directory for storing HDR images (e.g., ./data/HDRForCNN/), where the directory should have train, validation, and test directories.

4. Put HDR images into the train, validation, and test directories.

5. Rewrite the path to the data directory in ./conf/dataset/mydataset.yaml

6. Train iTM-Net. All outputs including trained models will be written in the history directory.

   poetry run python ./src/train.py
   

7. Test. All outputs will be written in the result directory.

   poetry run python ./src/test.py