CVPR 2022 | Dalian Institute of technology proposes a self calibration lighting framework for low light level image enhancement of real scenes

Thesis link ：

https://arxiv.org/pdf/2204.10137.pdf

Code link ：

https://github.com/vis-opt-group/SCI/blob/main/model.py

Abstract

The existing low light level image enhancement technology is not only difficult to deal with visual quality and computational efficiency , And it is usually invalid in unknown complex scenes . In this paper , We developed a New self calibration lighting （SCI） Learning framework , It is used to quickly 、 Flexibly and robustly illuminate images .

say concretely , We built a Cascade lighting learning process , And pass Weight sharing To handle this task . Considering the computational burden of cascade mode , We built Self calibration module , The convergence between the results of each stage is realized , Gain in reasoning using only a single basic block （ Not used in previous work ）, This greatly reduces the computational cost .

Then define Unsupervised training loss , To improve the ability of the model to adapt to general scenes .

Besides , We are right. SCI Inherent characteristics of （ There is a lack of ） Made a comprehensive exploration , Including operation insensitive adaptability （ Get stable performance under different simple operation settings ） Model independent commonality （ It can be applied to existing works based on lighting to improve performance ）.

Last , A large number of experiments and ablation studies have fully shown our advantages in quality and efficiency . The application in low light level face detection and night semantic segmentation fully reveals SCI The potential practical value of .