This is a pytorch project for the paper SNR-aware Low-Light Image Enhancement by Xiaogang Xu, Ruixing Wang, Chi-Wing Fu, and Jiaya Jia presented at CVPR2022.
This paper presents a new solution for low-light image enhancement by collectively exploiting Signal-to-Noise-Ratio-aware transformers and convolutional models to dynamically enhance pixels with spatial-varying operations. They are long-range operations for image regions of extremely low Signal-to-Noise-Ratio (SNR) and short-range operations for other regions. We propose to take an SNR prior to guide the feature fusion and formulate the SNR-aware transformer with a new selfattention model to avoid tokens from noisy image regions of very low SNR.
Please download the LOL-v1 and LOL-v2 from CVPR-2020-Semi-Low-Light.
The directory of LOL-v1 should contain our485 and eval15.
The directory of LOL-v2-real should contain Real_captured/Train and Real_captured/Test.
The directory of LOL-v2-synthetic should contain Synthetic/Train and Synthetic/Test.
Please place the directory of LOL-v1, LOL-v2-real, and LOL-v2-synthetic under "datasets".
For SID, we use the subset captured by the Sony camera and follow the script provided by SID to transfer the low-light images from RAW to RGB using rawpy’s default ISP.
For SMID, we use its full images and also transfer the RAWdata to RGB, since our work explores low-light image enhancement in the RGB domain.
You can download our processed datasets for SID and SMID from baidu pan (验证码: gplv) and baidu pan (验证码: btux), and there should contain "SMID_Long_np and SMID_LQ_np" and "long_sid2 and short_sid2". Moreover, the SMID dataset can also be downloaeded from google pan; the SID dataset can also be downloaded from google pan.
Different from original SDSD datasets with dynamic scenes, we utilize its static version (the scenes are the same of original SDSD).
And you can download the SDSD-indoor and SDSD-outdoor from baidu pan (验证码: jo1v) and baidu pan (验证码: uibk), and there should contain "indoor_static_np" and "outdoor_static_np". They can also be downloaded from google pan.
First install Python 3. We advise you to install Python 3 and PyTorch with Anaconda:
conda create --name py36 python=3.6
source activate py36
Clone the repo and install the complementary requirements:
cd $HOME
git clone --recursive git@github.com:dvlab-research/SNR-Low-Light-Enhance.git
cd SNR-Low-Light-Enhance
pip install -r requirements.txt
Train the model on the corresponding dataset using the train config. For example, the training on indoor subset of SDSD:
python -m torch.distributed.launch --nproc_per_node 1 --master_port 4320 train3.py -opt options/train/SDSD_indoor.yml --launcher pytorch
We use PSNR and SSIM as the metrics for evaluation. Evaluate the model on the corresponding dataset using the test config.
For the evaluation on indoor subset of SDSD, you should write the location of checkpoint in ``pretrain_model_G" of options/test/SDSD_indoor.yml use the following command line:
python test.py -opt options/test/SDSD_indoor.yml
For the evaluation on outdoor subset of SDSD, you should write the location of checkpoint in ``pretrain_model_G" of options/test/SDSD_outdoor.yml use the following command line:
python test.py -opt options/test/SDSD_outdoor.yml
For the evaluation on SMID, you should write the location of checkpoint in ``pretrain_model_G" of options/test/SMID.yml use the following command line:
python test.py -opt options/test/SMID.yml
For the evaluation on SID, you should write the location of checkpoint in ``pretrain_model_G" of options/test/SID.yml use the following command line:
python test.py -opt options/test/SID.yml
For the evaluation on LOL-v1, you should write the location of checkpoint in ``pretrain_model_G" of options/test/LOLv1.yml use the following command line:
python test_LOLv1_v2_real.py -opt options/test/LOLv1.yml
For the evaluation on LOL-v2-real, you should write the location of checkpoint in ``pretrain_model_G" of options/test/LOLv2_real.yml use the following command line:
python test_LOLv1_v2_real.py -opt options/test/LOLv2_real.yml
For the evaluation on LOL-v2-synthetic, you should write the location of checkpoint in ``pretrain_model_G" of options/test/LOLv2_synthetic.yml use the following command line:
python test_LOLv2_synthetic.py -opt options/test/LOLv2_synthetic.yml
You can download our trained model using the following links: https://drive.google.com/file/d/1g3NKmhz7WFLCm3t9qitqJqb_J7V4nzdb/view?usp=sharing
the model trained with indoor subset in SDSD: indoor_G.pth
the model trained with outdoor subset in SDSD: outdoor_G.pth
the model trained with SMID: smid_G.pth
the model trained with SID: sid_G.pth
the model trained with LOL-v1: LOLv1.pth
the model trained with LOL-v2-real: LOLv2_real.pth
the model trained with LOL-v2-synthetic: LOLv2_synthetic.pth
If you find the project useful, please cite:
@inproceedings{xu2022snr,
title={SNR-aware Low-Light Image Enhancement},
author={Xiaogang Xu, Ruixing Wang, Chi-Wing Fu, and Jiaya Jia},
booktitle={CVPR},
year={2022}
}
This source code is inspired by EDVR, attention-is-all-you-need-pytorch.
If you have any questions/comments/bug reports, feel free to e-mail the author Xiaogang Xu (xiaogangxu00@gmail.com).