An unofficial HDR project implementation of CVPR21 << A Two-stage Deep Network for High Dynamic Range Image Reconstruction>>
@inproceedings{a2021two,
title={A two-stage deep network for high dynamic range image reconstruction},
author={Sharif, SMA and Naqvi, Rizwan Ali and Biswas, Mithun and Kim, Sungjun}, booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition}, pages={550--559}, year={2021}}
official implementation code download:https://github.com/sharif-apu/twostageHDR_NTIRE21
paper download: https://arxiv.org/abs/2104.09386
My CSDN instruaction:https://blog.csdn.net/qq_41588833/article/details/119774573?spm=1001.2014.3001.5501
This is the official implementation of paper title "A Two-stage Deep Network for High Dynamic Range Image Reconstruction". The paper has been accepted and expected to be published in the proceedings of CVPRW21. To download full paper [Click Here].
Please consider to cite this paper as follows:
@inproceedings{a2021two,
title={A two-stage deep network for high dynamic range image reconstruction},
author={Sharif, SMA and Naqvi, Rizwan Ali and Biswas, Mithun and Kim, Sungjun},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={550--559},
year={2021}
}
Figure: Overview of the proposed method. The proposed method comprises a two-stage deep network. Stage-I aims toperform image enhancement task denoising, exposure correction, etc. Stage-II of the proposed method intends to performtone mapping and bit-expansion.
Figure: Quantitative comparison between proposed method and existing learning-based single-shot LDR to HDR methods..
Python 3.8
CUDA 10.1 + CuDNN
pip
Virtual environment (optional)
Please consider using a virtual environment to continue the installation process.
git clone https://github.com/sharif-apu/twostageHDR_NTIRE21.git
cd twostageHDR_NTIRE21
pip install -r requirement.txt
To download the training images please visit the following link: [Click Here] and extract the zip files in common directory.
The original paper used image patches from HdM HDR dataset. To extract image patches please execute Extras/processHDMDHR.py script from the root directory as follows:
python processHDMDHR.py -r path/to/HdM/root/ -t path/to/save/patch -p 256
Here -r flag defines your root directory of the HdM HDR training samples, -s flag defines the directory where patches should be saved, and -p flag defines the patch size
After extracting patch please execute the following commands to start training:
python main.py -ts -e X -b Y
To specify your trining images path, go to mainModule/config.json and update "trainingImagePath" entity.
You can specify the number of epoch with -e flag (i.e., -e 5) and number of images per batch with -b flag (i.e., -b 24).
Please Note: The provided code aims to train only with medium exposure frames. To train with short/long exposure frames, you need to modify the dataTools/customDataloader (line 68) and mainModule/twostageHDR (line 87).
For transfer learning execute:
python main.py -tr -e -b
The provided weights are trained as per rule of NTIRE21 HDR challange (single frame). To download the testing images please visit the following link: [Click Here]
To inference with provided pretrained weights please execute the following commands:
python main.py -i -s path/to/inputImages -d path/to/outputImages
Here,-s specifies the root directory of the source images
(i.e., testingImages/), and -d specifies the destination root (i.e., modelOutput/).
We have collected a LDR dataset captured with different camera hardwares. Feel free to use our LDR dataset in your respective work. The dataset can be downloaded from the following link: [Click Here]
For any further query, feel free to contact us through the following emails: sma.sharif.cse@ulab.edu.bd, rizwanali@sejong.ac.kr, or mithun.bishwash.cse@ulab.edu.bd