Figure: Architecture of the proposed Improved Residual Network (IRNet) and the Improved Residual Block (IRB)
Clone this github repository
git clone https://github.com/ThisisVikki/ITM-baseline.git
cd ITM-baseline
Create a new environment and install the dependencies
conda create -n IRNet python=3.8 -y
conda activate IRNet
pip install -r requirements.txt
We use the HDRTV1K dataset for both training and testing. The test set of Deep SR-ITM and our ITM-4K test dataset are also used for testing. The ITM-4K test dataset can be downloaded from Google Drive and Baidu NetDisk (access code: t7iy). It contains 160 pairs of SDR and HDR images of size
The model is the model that needs to be tested, e.g. IRNet_2, IRNet_1 or SRITM_IRNet_5. Please make sure the path of the test data testdata_path and pretrained model model_path in photo.py are correct, then run the code:
python photo.py
The test result will be saved to ./results.
The pretrained model will be released soon.
The training settings can be found at ./experiments/IRNet_COSINE.yaml, please check the settings and make sure TRAIN_DATAROOT_GT, TRAIN_DATAROOT_LQ, VALID_DATAROOT_GT and VALID_DATAROOT_LQ are in the right paths.
Then run the code:
python train.py --model [model name] --channels [model_channels]
If our work is helpful for you, please cite our paper:
@misc{xue2023lightweight,
title={Lightweight Improved Residual Network for Efficient Inverse Tone Mapping},
author={Liqi Xue and Tianyi Xu and Yongbao Song and Yan Liu and Lei Zhang and Xiantong Zhen and Jun Xu},
year={2023},
eprint={2307.03998},
archivePrefix={arXiv},
primaryClass={cs.CV}
}The code and the proposed test dataset is released for academic research only.