PMambaIR：panoramic vision state space model for lightweight image super-resolution

Gang YAN; Ziyi SONG; Shuze GENG

doi:10.37188/CJLCD.2024-0246

Chinese Journal of Liquid Crystals and Displays, Volume. 40, Issue 4, 642(2025)

PMambaIR：panoramic vision state space model for lightweight image super-resolution

Gang YAN¹, Ziyi SONG¹, and Shuze GENG^2、*

Author Affiliations

¹School of Artificial Intelligence，Hebei University of Technology，Tianjin 300131，China

²School of Information Science and Technology，Tianjin University of Technology and Education，Tianjin 300222，China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Overall framework of PMambaIR.（a）Multi-scale cascade scan（MSCS）；（b）Hybrid state space block（HSSB）.

Download full size

View in Article

Fig. 2. Multi-scale cascade scan，displaying horizontal forward scanning.（a）Local level；（b）Middle level；（c）Global level.

Download full size

View in Article

Fig. 3. Visual comparison of the methods for ×4 SR，the comparison area is marked with a red box.

Download full size

View in Article

Fig. 4. Comparion of LAM visualization of PMambaIR with other based methods on ×4 SR

Download full size

View in Article

Fig. 5. Qualitative comparison on models with different window sizes.（a）Window size 4×4；（b）Window size 8×8；（c）Window size 16×16.

Download full size

View in Article

Table 1. Quantitative comparison of method in this paper with state-of-the-art methods on benchmark datasets

View table

View in Article

Table 1. Quantitative comparison of method in this paper with state-of-the-art methods on benchmark datasets

方法	比例	参数/K	MACs/G	Set5		Set14		BSDS100		Urban100		Manga109
方法	比例	参数/K	MACs/G	PSNR	SSIM	PSNR	SSIM	PSNR	SSIM	PSNR	SSIM	PSNR	SSIM
EDSR ^［4］	$\times$ 2	1 370	316	37.99	0.960 4	33.57	0.917 5	32.16	0.899 4	31.98	0.927 2	38.54	0.976 9
CARN ^［5］	$\times$ 2	1 592	222	37.76	0.959 0	33.52	0.916 6	32.09	0.897 8	31.92	0.925 6	38.36	0.976 5
IMDN ^［6］	$\times$ 2	694	158	28.00	0.960 5	33.63	0.917 7	32.19	0.899 6	32.17	0.928 3	38.88	0.977 4
LatticeNet ^［8］	$\times$ 2	756	169	38.15	0.961 0	33.78	0.919 3	32.25	0.900 5	32.43	0.930 2	-	-
SwinIR ^［1］	$\times$ 2	878	195	38.14	0.961 1	33.86	0.920 7	32.31	0.901 2	32.76	0.934 0	39.12	0.978 3
Omni-SR ^［2］	$\times$ 2	772	-	38.22	0.961 3	33.98	0.921 0	33.36	0.902 0	33.05	0.936 3	39.28	0.978 4
SRFormer ^［3］	$\times$ 2	853	236	38.23	0.961 3	33.94	0.920 9	32.36	0.901 9	32.91	0.935 3	39.28	0.978 5
MambaIR ^［10］	$\times$ 2	859	198	38.16	0.961 0	34.00	0.921 2	32.34	0.901 7	32.92	0.935 6	39.31	0.977 9
Ours	$\times$ 2	970	201	38.23	0.961 8	34.04	0.922 1	33.37	0.902 3	33.03	0.935 9	39.53	0.978 8
EDSR ^［4］	$\times$ 3	1 555	160	34.37	0.927 0	30.28	0.841 7	29.09	0.805 2	28.15	0.852 7	33.45	0.943 9
CARN ^［5］	$\times$ 3	1 592	118	34.29	0.925 5	30.29	0.840 7	29.06	0.803 4	28.06	0.849 3	33.50	0.944 0
IMDN ^［6］	$\times$ 3	703	71.5	34.36	0.927 0	30.32	0.804 6	29.09	0.804 6	28.17	0.851 9	33.61	0.944 5
LatticeNet ^［8］	$\times$ 3	765	76.3	34.53	0.928 1	30.39	0.842 4	29.15	0.805 9	28.33	0.853 8	-	-
ESRT ^［7］	$\times$ 3	770	-	34.42	0.926 8	30.43	0.843 3	29.15	0.806 3	28.46	0.857 4	33.95	0.945 5
SwinIR ^［1］	$\times$ 3	868	87.2	34.62	0.928 9	30.54	0.846 3	29.20	0.808 2	28.66	0.862 4	33.98	0.947 9
Omni-SR ^［2］	$\times$ 3	780	-	34.70	0.929 4	30.57	0.846 9	29.28	0.809 4	28.84	0.865 6	34.22	0.948 7
SRFormer ^［3］	$\times$ 3	861	105	34.67	0.929 6	30.57	0.846 9	29.26	0.809 9	28.81	0.865 5	34.19	0.948 9
MambaIR ^［10］	$\times$ 3	867	88.7	34.72	0.929 6	30.63	0.847 5	29.29	0.809 9	29.00	0.868 9	34.39	0.949 5
Ours	$\times$ 3	970	90.1	34.75	0.930 5	30.66	0.848 8	29.31	0.810 4	29.12	0.869 7	34.65	0.950 4
EDSR ^［4］	$\times$ 4	1 518	114	32.09	0.893 8	28.58	0.781 3	27.57	0.735 7	26.06	0.784 9	30.25	0.906 7
CARN ^［5］	$\times$ 4	1 592	90.9	32.13	0.893 7	28.60	0.780 6	27.58	0.734 9	26.07	0.783 7	30.47	0.908 4
IMDN ^［6］	$\times$ 4	715	40.9	32.21	0.894 8	28.58	0.781 1	27.56	0.735 3	26.04	0.783 8	30.45	0.907 5
LatticeNet ^［8］	$\times$ 4	777	43.6	32.30	0.896 2	28.68	0.783 0	27.62	0.736 7	26.25	0.787 3	-	-
ESRT ^［7］	$\times$ 4	751	-	32.19	0.894 7	28.69	0.783 3	27.69	0.737 9	26.39	0.796 3	30.75	0.910 0
SwinIR ^［1］	$\times$ 4	897	49.6	32.44	0.897 6	28.77	0.785 8	27.69	0.740 6	26.47	0.798 0	30.92	0.915 1
Omni-SR ^［2］	$\times$ 4	792	46.5	32.49	0.898 8	28.78	0.785 9	27.71	0.741 5	26.64	0.801 8	31.02	0.915 1
SRFormer ^［3］	$\times$ 4	873	62.8	32.51	0.898 8	28.82	8.787 2	27.73	0.742 2	26.67	0.803 2	31.17	0.916 5
MambaIR ^［10］	$\times$ 4	879	50.6	32.51	0.899 3	28.85	0.787 2	27.75	0.742 3	26.75	0.805 1	31.26	0.917 5
Ours	$\times$ 4	970	45.5	32.55	0.900 1	28.88	0.789 4	27.76	0.743 6	26.87	0.807 5	31.37	0.918 1

Table 2. Quantitative results of models with different window sizes for ×4 SR

View table

View in Article

Table 2. Quantitative results of models with different window sizes for ×4 SR

窗口大小	Set14	BSDS100	Urban100	Manga109
窗口大小	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM	PSNR/SSIM
（4×4）	28.79/0.786 5	27.70/0.740 7	26.56/0.799 8	31.06/0.913 8
（8×8）	28.85/0.787 6	27.74/0.741 8	26.72/0.804 0	31.21/0.916 6
（16×16）	28.82/0.787 0	27.74/0.741 7	26.71/0.807 5	31.15/0.916 2

Table 3. Ablation experiment on the effectiveness of HSSB

View table

View in Article

Table 3. Ablation experiment on the effectiveness of HSSB

Exp.	CAB	MLP	Res.	Prams/K	Set14	Urban100
Exp.	CAB	MLP	Res.	Prams/K	PSNR/SSIM	PSNR/SSIM
1	×	×	×	612	28.76/0.785 7	26.58/0.799 1
2	√	×	×	880	28.81/0.787 4	26.67/0.802 4
3	×	√	√	702	28.79/0.786 7	26.65/0.801 8
4	√	×	√	880	28.85/0.788 6	26.79/0.805 9
5	√	√	√	970	28.88/0.789 4	26.87/0.807 5

Table 4. Ablation experiment of HSSB quantity

View table

View in Article

Table 4. Ablation experiment of HSSB quantity

Exp.	Params/K	HSSB number	Set14	Urban100
Exp.	Params/K	HSSB number	PSNR/SSIM	PSNR/SSIM
1	580.5	3，3，3，3	28.69/0.783 4	26.40/0.796 5
2	710.7	4，4，4，4	28.76/0.785 8	26.64/0.801 1
3	970.2	6，6，6，6	28.88/0.789 4	26.87/0.807 5
4	1 231.2	8，8，8，8	28.90/0.789 8	26.89/0.809 0

Table 5. Ablation experiments using different scanning methods in HSSB

View table

View in Article

Table 5. Ablation experiments using different scanning methods in HSSB

Exp.	Window level	Middle level	Global level	Set14	Urban100
Exp.	Window level	Middle level	Global level	PSNR/SSIM	PSNR/SSIM
1	√	×	×	28.71/0.783 9	26.46/0.797 6
2	×	√	×	28.74 /0.785 0	26.58/0.799 6
3	×	×	√	28.81/0.786 9	26.69/0.803 1
4	√	√	×	28.82/0.787 2	26.73/0.804 6
5	√	×	√	28.82/0.787 4	26.74/0.804 9
6	×	√	√	28.85/0.788 7	26.82/0.806 2
7	√	√	√	28.88/0.789 4	26.87/0.807 5

Tools

Get Citation

Copy Citation Text

Gang YAN, Ziyi SONG, Shuze GENG. PMambaIR：panoramic vision state space model for lightweight image super-resolution[J]. Chinese Journal of Liquid Crystals and Displays, 2025, 40(4): 642

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category:

Received: Aug. 22, 2024

Accepted: --

Published Online: May. 21, 2025

The Author Email: Shuze GENG (gengshuze@tute.edu.cn)

DOI:10.37188/CJLCD.2024-0246

Topics