notesum.ai

Published at December 5

LocalSR: Image Super-Resolution in Local Region

cs.CV

Released Date: December 5, 2024

Authors: Bo Ji¹, Angela Yao¹

Aff.: ¹National University of Singapore

Arxiv: http://arxiv.org/pdf/2412.04314v1

Refer to caption

	Method	Variant	BSD100			Urban100			Manga109			Test2K
	Method	Variant	PSNR	SSIM	GFLOPs	PSNR	SSIM	GFLOPs	PSNR	SSIM	GFLOPs	PSNR	SSIM	GFLOPs
CNN-based	CARN [2]	Pre-cropping	27.27	0.8839	0.65	25.09	0.8852	0.65	28.82	0.9574	0.65	27.23	0.8893	0.65
		Post-cropping	27.38	0.8854	35.37	25.29	0.8879	209.59	29.27	0.9597	267.23	27.34	0.8910	489.93
		Ours	27.39	0.8854	1.63	25.32	0.8883	5.65	29.24	0.9595	6.97	27.34	0.8909	12.11
	SRResNet [15]	Pre-cropping	27.51	0.8880	2.93	25.77	0.8983	2.93	29.74	0.9640	2.93	27.44	0.8939	2.93
		Post-cropping	27.63	0.8895	158.19	25.99	0.9013	937.44	30.19	0.9658	1195.23	27.56	0.8957	2191.26
		Ours	27.62	0.8894	5.08	26.03	0.9018	8.94	30.20	0.9656	10.22	27.57	0.8957	15.15
	RCAN [36]	Pre-cropping	27.67	0.8909	18.38	26.35	0.9087	18.38	30.47	0.9682	18.38	27.60	0.8971	18.38
		Post-cropping	27.82	0.8926	992.73	26.67	0.9126	5882.85	31.07	0.9707	7500.64	27.76	0.8991	13751.17
		Ours	27.79	0.8924	19.41	26.66	0.9124	21.50	31.05	0.9706	22.19	27.75	0.8990	24.86
Transformer-based	SwinIR (lw) [16]	Pre-cropping	27.63	0.8901	2.11	26.09	0.9045	2.11	30.26	0.9673	2.11	27.56	0.8962	2.11
		Post-cropping	27.76	0.8917	113.70	26.37	0.9082	673.77	30.79	0.9693	859.06	27.71	0.8982	1574.94
		Ours	27.76	0.8918	3.18	26.38	0.9084	6.74	30.80	0.9692	7.92	27.71	0.8982	12.47
	SwinIR [16]	Pre-cropping	27.75	0.8923	15.08	26.53	0.9118	15.08	30.76	0.9706	15.08	27.68	0.8987	15.08
		Post-cropping	27.90	0.8942	814.40	26.90	0.9165	4826.06	31.36	0.9728	6153.23	27.86	0.9011	11280.91
		Ours	27.90	0.8942	23.23	26.91	0.9165	31.02	31.36	0.9728	33.60	27.85	0.9010	43.56