notesum.ai

Published at November 27

Complexity Experts are Task-Discriminative Learners for Any Image Restoration

cs.CV

Released Date: November 27, 2024

Authors: Eduard Zamfir¹, Zongwei Wu¹, Nancy Mehta¹, Yuedong Tan¹, Danda Pani Paudel², Yulun Zhang³, Radu Timofte¹

Aff.: ¹University of Wurzburg; ²INSAIT Sofia University; ³Shanghai Jiao Tong University

Arxiv: http://arxiv.org/abs/2411.18466v1

Refer to caption

	Method	Params.	Dehazing		Deraining		Denoising						Average
	Method	Params.	SOTS		Rain100L		BSD68_{$\sigma$ =15}		BSD68_{$\sigma$ =25}		BSD68_{$\sigma$ =50}		Average
Light	BRDNet [48]	-	23.23	.895	27.42	.895	32.26	.898	29.76	.836	26.34	.693	27.80	.843
	LPNet [14]	-	20.84	828	24.88	.784	26.47	.778	24.77	.748	21.26	.552	23.64	.738
	FDGAN [11]	-	24.71	.929	29.89	.933	30.25	.910	28.81	.868	26.43	.776	28.02	.883
	DL [13]	2M	26.92	.931	32.62	.931	33.05	.914	30.41	.861	26.90	.740	29.98	.876
	AirNet [24]	9M	27.94	.962	34.90	.967	33.92	.933	31.26	.888	28.00	.797	31.20	.910
	MoCE-IR-S (ours)	11M	30.94	.979	38.22	.983	34.08	.933	31.42	.888	28.16	.798	32.57	.916
Heavy	MPRNet [63]	16M	25.28	.955	33.57	.954	33.54	.927	30.89	.880	27.56	.779	30.17	.899
	PromptIR [36]	36M	30.58	.974	36.37	.972	33.98	.933	31.31	.888	28.06	.799	32.06	.913
	Gridformer [51]	34M	30.37	.970	37.15	.972	33.93	.931	31.37	.887	28.11	.801	32.19	.912
	Art-PromptIR [54]	33M	30.83	.979	37.94	.982	34.06	.934	31.42	.891	28.14	.801	32.49	.917
	DA-CLIP^∗ [30]	125M	29.46	.963	36.28	.968	30.02	.821	24.86	.585	22.29	.476	-	-
	UniProcessor^∗ [12]	1002M	31.66	.979	38.17	.982	34.08	.935	31.42	.891	28.17	.803	32.70	.918
	MoCE-IR (ours)	25M	31.12	.979	38.72	.984	34.13	.932	31.47	.888	28.20	.800	32.73	.917