Cross-level feature adaptive fusion network for low-light image enhancement

Method	PSNR↑	SSIM↑	LPIPS↓	MSE↓
BPDHE^［2］	12.22	0.344	0.547	0.084 8
LIME^［5］	14.15	0.366	0.500	0.042 9
Dong^［6］	15.96	0.541	0.330	0.044 8
ZERO-DCE^［10］	15.58	0.511	0.431	0.036 37
KinD^［9］	19.29	0.805	0.207	0.016 4
URetinexNet^［14］	21.33	0.833	0.121	0.010 50
TreEnhance^［13］	21.96	0.81	-	-
DCCNet^［24］	22.72	0.81	0.143	0.010 08
SurroundNet^［12］	22.81	0.853	0.190	-
LLformer^［25］	23.65	0.816	0.169	0.007 71
Ours	23.65	0.857	0.113	0.008 31

Table 2. Experimental results on the LOLv2 dataset

View table

View in Article

Table 2. Experimental results on the LOLv2 dataset

Method	PSNR↑	SSIM↑	LPIPS↓	MSE↓
BPDHE	13.77	0.389	0.458	0.050 35
LIME	16.95	0.535	0.348	0.030 33
Dong	17.90	0.546	0.335	0.026 61
ZERO-DCE	12.00	0.438	0.510	0.076 8
KinD	17.58	0.735	0.448	0.021 14
LLformer	20.23	0.782	0.261	0.013 25
DCCNet	20.49	0.814	0.220	0.012 88
SurroundNet	21.30	0.817	0.200	0.010 11
Ours	21.17	0.853	0.161	0.012 61

Table 3. Block ablation results
View table
View in Article
Table 3. Block ablation results
PPM MTAM SKFF PSNR↑ SSIM↑ LPIPS↓ MSE↓
M1 √ √ 22.36 0.841 0.125 0.011 43
M2 √ √ 22.14 0.833 0.143 0.009 79
M3 √ √ 22.92 0.850 0.122 0.009 07
M1+M2+M3 √ √ √ 23.65 0.857 0.113 0.008 31

Table 4. Loss function ablation results
View table
View in Article
Table 4. Loss function ablation results
CharLoss EdgeLoss SSIMLoss PSNR↑ SSIM↑ LPIPS↓ MSE↓
M1 √ √ 21.88 0.852 0.121 0.011 44
M2 √ √ 23.16 0.855 0.117 0.009 08
M3 √ √ 23.45 0.844 0.119 0.008 44
M1+M2+M3 √ √ √ 23.65 0.857 0.113 0.008 31

Tools

Get Citation

Copy Citation Text

Liming LIANG, Chenkun ZHU, Yuan YANG, Renjie LI. Cross-level feature adaptive fusion network for low-light image enhancement[J]. Chinese Journal of Liquid Crystals and Displays, 2024, 39(6): 856

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites