Table 11 Quantitative results on the darkest \(10\%\) images (worst-case subset) from the LOL and LOL-v2 datasets. Higher is better for \(\uparrow\) metrics; lower is better for \(\downarrow\) metrics. Best performance per metric is highlighted in bold.
From: A multi scale spatial attention based zero shot learning framework for low light image enhancement
Metric | Paired supervision | Unsupervised | Zero-shot learning | Ours | |||||
|---|---|---|---|---|---|---|---|---|---|
BIMEF | LIME | MF | MSR | EnlightenGAN | SG-ZSL | Zero-DCE | Zero-DCE++ | ||
PSNR (\(\uparrow\)) | 18.46 | 16.15 | 20.12 | 12.18 | 19.42 | 19.31 | 18.63 | 21.18 | 21.26 |
SSIM (\(\uparrow\)) | 0.82 | 0.71 | 0.83 | 0.63 | 0.82 | 0.84 | 0.83 | 0.84 | 0.86 |
FSIM (\(\uparrow\)) | 0.94 | 0.80 | 0.90 | 0.77 | 0.92 | 0.94 | 0.89 | 0.92 | 0.95 |
VSI (\(\uparrow\)) | 0.98 | 0.93 | 0.97 | 0.91 | 0.97 | 0.98 | 0.96 | 0.97 | 0.99 |
LPIPS (\(\downarrow\)) | 0.14 | 0.24 | 0.20 | 0.27 | 0.14 | 0.15 | 0.20 | 0.15 | 0.10 |
DISTS (\(\downarrow\)) | 0.15 | 0.19 | 0.16 | 0.19 | 0.13 | 0.14 | 0.16 | 0.14 | 0.12 |
MAD (\(\downarrow\)) | 112.64 | 160.03 | 133.89 | 166.87 | 115.56 | 115.57 | 134.12 | 119.03 | 110.74 |
