Official PyTorch implementation of "Single Image HDR Reconstruction Using a CNN with Masked Features and Perceptual Loss" (SIGGRAPH 2020) Project | Paper
We propose a novel deep learning approach to reconstruct an HDR image by recovering the saturated pixels of a single input LDR image in a visually pleasing way. Our method can reconstruct regions with high luminance, such as the bright highlights of the windows (red inset), and generate visually pleasing textures and details (green insert). For more information on the method please see the project website.
This codebase was developed and tested with PyTorch 1.2 and Python 3.6.
- Python 3.6+
- Pytorch 1.2
- torchvision
- OpenCV
- Numpy
- tensorboardX
- tqdm
- Pillow
- pyexr
- OpenEXR
pip install -r requirements.txt
You may have to install OpenEXR through the appropriate package manager before pip install (e.g. sudo apt-get install openexr and libopenexr-dev on Ubuntu).
Download the repository
https://github.com/marcelsan/Deep-HdrReconstruction.git
The pretrained model checkpoints can be found in the checkpoints folder on Google Drive.
CUDA_VISIBLE_DEVICES=1 python test_hdr.py --test_dir <images/dir> --out_dir <out/dir> --weights <weight/path>.pth
Parameters and their description:
test_dir: input images directory. A few images are avaible on the data/ folder.
out_dir: path to output directory.
weights: path to the trained CNN weights.
If cuda is available, it will be used. In case you want to run the model on cpu, use --cpu when executing test_hdr.py
If you find this work useful for your research, please cite:
@article{Marcel:2020:LDRHDR,
author = {Santos, Marcel Santana and Tsang, Ren and Khademi Kalantari, Nima},
title = {Single Image HDR Reconstruction Using a CNN with Masked Features and Perceptual Loss},
journal = {ACM Transactions on Graphics},
volume = {39},
number = {4},
year = {2020},
month = {7},
doi = {10.1145/3386569.3392403}
}
Please contact Marcel Santos (mss8@cin.ufpe.br) if there are any issues/comments/questions.
Copyright (c) 2020, Marcel Santana.
All rights reserved.
The code is distributed under a BSD license. See LICENSE for information.