notesum.ai

Published at November 26

MWFormer: Multi-Weather Image Restoration Using Degradation-Aware Transformers

cs.CV

Released Date: November 26, 2024

Authors: Ruoxi Zhu¹, Zhengzhong Tu¹, Jiaming Liu¹, Alan C. Bovik¹, Yibo Fan¹

Aff.: ¹Unknown

Arxiv: http://arxiv.org/abs/2411.17226v1

Refer to caption

Model	RainDrop [31]		Outdoor-Rain [50]		Snow100K [51]		Average		MACs (G) $\downarrow$
Model	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	PSNR $\uparrow$	SSIM $\uparrow$	MACs (G) $\downarrow$
AirNet[13]	24.57	0.8583	18.48	0.6719	24.41	0.8045	22.49	0.7782	301.27
Chen et al.[36]	31.83	0.9289	25.45	0.8737	28.86	0.8886	28.71	0.8971	24.56
All-in-One[12]	31.12	0.9268	24.71	0.8980	28.33	0.8820	28.05	0.9023	12.26
TransWeather[2]	30.17	0.9157	28.83	0.9000	29.31	0.8879	29.44	0.9012	6.13
WeatherDiffusion[1]	30.71	0.9312	29.64	0.9312	30.09	0.9041	30.15	0.9222	475.16 $\times$ 50
Zhu et al.[37]	31.31	0.93	25.31	0.90	29.71	0.89	28.78	0.91	1.36
MWFormer-S (ours)	31.09	0.9224	29.07	0.9010	30.05	0.8986	30.07	0.9073	3.57
MWFormer-M (ours)	31.56	0.9246	29.70	0.9064	30.45	0.9029	30.57	0.9113	6.45
MWFormer-L (ours)	31.73	0.9254	30.24	0.9111	30.70	0.9060	30.89	0.9142	10.41
MWFormer-real^∗ (ours)	31.91	0.9268	30.27	0.9121	30.92	0.9084	31.03	0.9158	10.41