
摘要
在增强低光图像时,许多深度学习算法都基于Retinex理论。然而,Retinex模型并未考虑暗部隐藏的噪声或由增亮过程引入的噪声。此外,这些方法通常需要繁琐的多阶段训练流程,并依赖于卷积神经网络,这在捕捉长距离依赖关系方面显示出局限性。本文提出了一种简单而有原则的一阶段Retinex基础框架(One-stage Retinex-based Framework,简称ORF)。ORF首先估计光照信息以照亮低光图像,然后恢复图像中的噪声以生成增强后的图像。我们设计了一种光照引导的Transformer(Illumination-Guided Transformer,简称IGT),该模型利用光照表示来指导不同光照条件下区域之间的非局部相互作用建模。通过将IGT集成到ORF中,我们得到了我们的算法——Retinexformer。全面的定量和定性实验表明,我们的Retinexformer在十三个基准测试中显著优于现有最先进方法。用户研究和低光物体检测应用也揭示了我们方法潜在的实际价值。代码、模型和结果可在https://github.com/caiyuanhao1998/Retinexformer 获取。
代码仓库
caiyuanhao1998/retinexformer
官方
pytorch
GitHub 中提及
dawnlh/awesome-low-light-image-enhancement
pytorch
GitHub 中提及
基准测试
| 基准 | 方法 | 指标 |
|---|---|---|
| image-enhancement-on-mit-adobe-5k | Retinexformer | PSNR on proRGB: 25.98 PSNR on sRGB: 24.94 SSIM on proRGB: 0.957 SSIM on sRGB: 0.907 |
| low-light-image-deblurring-and-enhancement-on | RetinexFormer | Average PSNR: 22.904 LPIPS: 0.236 SSIM: 0.824 |
| low-light-image-enhancement-on-dicm | Retinexformer | User Study Score: 3.71 |
| low-light-image-enhancement-on-lime | Rextinexformer | User Study Score: 4.3 |
| low-light-image-enhancement-on-lol | Retinexformer_ | Average PSNR: 27.18 FLOPS (G): 15.57 Params (M): 1.61 SSIM: 0.850 |
| low-light-image-enhancement-on-lol | Retinexformer | Average PSNR: 25.16 FLOPS (G): 15.57 Params (M): 1.61 SSIM: 0.845 |
| low-light-image-enhancement-on-lol-v2 | Retinexformer | Average PSNR: 22.80 SSIM: 0.840 |
| low-light-image-enhancement-on-lol-v2-1 | Retinexformer | PSNR: 25.67 SSIM: 0.939 |
| low-light-image-enhancement-on-lolv2 | Retinexformer | Average PSNR: 27.71 SSIM: 0.856 |
| low-light-image-enhancement-on-lolv2-1 | Retinexformer | Average PSNR: 29.04 SSIM: 0.939 |
| low-light-image-enhancement-on-mef | Retinexformer | User Study Score: 3.91 |
| low-light-image-enhancement-on-mit-adobe-1 | Retinexformer | PSNR: 24.94 SSIM: 0.907 |
| low-light-image-enhancement-on-npe | Retinexformer | User Study Score: 4.17 |
| low-light-image-enhancement-on-sdsd-indoor | Retinexformer | PSNR: 29.77 |
| low-light-image-enhancement-on-sdsd-outdoor | Retinexformer | PSNR: 29.84 |
| low-light-image-enhancement-on-sid | Retinexformer | PSNR: 24.44 SSIM: 0.680 |
| low-light-image-enhancement-on-smid | Retinexformer | PSNR: 29.15 |
| low-light-image-enhancement-on-vv | Rextinexformer | User Study Score: 3.61 |
| photo-retouching-on-mit-adobe-5k | Retinexformer | PSNR: 24.94 SSIM: 0.907 |