This repository contains the code for the paper:
MLP Embedded Inverse Tone Mapping, accepted at MM2024 (ACM International Conference on Multimedia 2024).
This project implements a novel approach to inverse tone mapping using an MLP-based architecture. The code is designed for training and testing on HDRTV4K datasets, but it can be adapted to other datasets as well.
-
Clone the repository:
git clone https://github.com/pjliu3/MLP_iTM.git cd MLP_iTM -
Install the required dependencies:
pip install -r requirements.txt
We provide a pretrained model on the HDRTV4K dataset, which you can find in the pre_model folder. You can directly use this model for testing if desired.
To train the model, you need to organize your dataset as follows:
- Place the HDR images in
dataset/train/hdr. - Place the SDR images in
dataset/train/sdr.
Follow the format of the provided JSON example to create your own JSON file that describes the dataset. The JSON file should contain the paths to the paired HDR and SDR images for training.
["abp1_autumnwoods_000.jpg", "abp1_bamboo_000.jpg", "uzbek_044.jpg", "uzbek_054.jpg", "uzbek_064.jpg", "venice_006.jpg", "venice_016.jpg", "winter_002.jpg", "winter_012.jpg",]
To start pretraining, ensure your dataset is correctly set up, then run:
python pretrain.pyYou can use the pretrained model for testing on new images. First, ensure that your test images are placed in the dataset/test folder. Then, run the following command:
python test_ours.pyThe output results will be saved in the ./results directory.
- Pretraining: We used 8 NVIDIA RTX 3090 GPUs for pretraining.
- Inference: Inference was tested on a single NVIDIA RTX 4090 GPU.
If you have any questions or encounter any issues, feel free to contact us at:
If you find this code useful in your research, please cite our paper:
@inproceedings{Liu2024MLP,
author = {Panjun Liu and Jiacheng Li and Lizhi Wang and Zheng-Jun Zha and Zhiwei Xiong},
title = {MLP Embedded Inverse Tone Mapping},
booktitle = {Proceedings of the 32nd ACM International Conference on Multimedia (MM '24)},
year = {2024},
location = {Melbourne, VIC, Australia},
pages = {9 pages},
publisher = {ACM},
address = {New York, NY, USA},
doi = {https://doi.org/10.1145/3664647.3680937},
month = {October 28-November 1}
}
This project is licensed under the MIT License - see the LICENSE file for details.