Awesome

UTVNet: Adaptive Unfolding Total Variation Network for Low-Light Image Enhancement

This repo is the official implementation of ICCV2021 paper "Adaptive Unfolding Total Variation Network for Low-Light Image Enhancement". For more details, please see our paper.

UTVNet

Real-world low-light images suffer from two main degradations, namely, inevitable noise and poor visibility. Since the noise exhibits different levels, its estimation has been implemented in recent works when enhancing low-light images from raw Bayer space. When it comes to sRGB color space, the noise estimation becomes more complicated due to the effect of the image processing pipeline. Nevertheless, most existing enhancing algorithms in sRGB space only focus on the low visibility problem or suppress the noise under a hypothetical noise level, leading them impractical due to the lack of robustness. To address this issue, we propose an adaptive unfolding total variation network (UTVNet), which approximates the noise level from the real sRGB low-light image by learning the balancing parameter in the model-based denoising method with total variation regularization. Meanwhile, we learn the noise level map by unrolling the corresponding minimization process for providing the inferences of smoothness and fidelity constraints. Guided by the noise level map, our UTVNet can recover finer details and is more capable to suppress noise in real captured low-light scenes. Extensive experiments on real-world low-light images clearly demonstrate the superior performance of UTVNet over state-of-the-art methods.

Code

Prerequisites

• Python 3.6 and Pytorch 1.7.1.
• Requirements: Numpy, skiamge, tqdm, opencv-python, matplotlib, pillow

Data

ELD dataset

• ELD dataset website,
• Download the selected sRGB version used in our model website
• Copy the dataset to dataset/ELD

sRGBSID

• Contact the author by website.
• Copy the Test and Ground-Truth dataset to dataset/sRGBSID

Model

Download the pretrained model from website, copy the model to pretrain_model

Evaluation

sRGBSID

python test.py --data_name='sRGBSID'
python evaluate.py --data_name='sRGBSID'

please check the results in result/sRGBSID

ELD

• Sony

python test.py --data_name='ELD_sony'
python evaluate.py --data_name='ELD_sony'

please check the results in result/ELD_sony

• Cano

python test.py --data_name='ELD_cano'
python evaluate.py --data_name='ELD_cano'

please check the results in result/ELD_cano

• Niko

python test.py --data_name='ELD_niko'
python evaluate.py --data_name='ELD_niko'

please check the results in result/ELD_niko

Test Result

If you find it difficult to generate the test results, plesase send me an email.(chuanjunzhengcs@gmail.com)

Citation

If you find our work useful in your research, please consider citing:

@InProceedings{Zheng_2021_ICCV,
    author    = {Zheng, Chuanjun and Shi, Daming and Shi, Wentian},
    title     = {Adaptive Unfolding Total Variation Network for Low-Light Image Enhancement},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2021},
    pages     = {4439-4448}
}

Acknowledgements

The code in basicblock.py is based on Model USRNet