RASLF: Representation-Aware State Space Model for Light Field Super-Resolution

Zeqiang Wei; Kai Jin; Kuan Song; Xiuzhuang Zhou; Wenlong Chen; Min Xu

cs.CVMar 17, 2026

RASLF: Representation-Aware State Space Model for Light Field Super-Resolution

Authors: Zeqiang Wei, Kai Jin, Kuan Song, Xiuzhuang Zhou, Wenlong Chen, Min Xu

Organizations: Capital Normal University Information Engineering College, Beijing 100048, China · Bigo Technology Pte. Ltd., Beijing 100020, China · Explorer Global (Suzhou) Artificial Intelligence Technology Co., Ltd., Suzhou, Jiangsu 215123, China · Beijing University of Posts and Telecommunications, Beijing 100088, China

Abstract

Current SSM-based light field super-resolution (LFSR) methods often fail to fully leverage the complementarity among various LF representations, leading to the loss of fine textures and geometric misalignments across views. To address these issues, we propose RASLF, a representation-aware state-space framework that explicitly models structural correlations across multiple LF representations. Specifically, a Progressive Geometric Refinement (PGR) block is created that uses a panoramic epipolar representation to explicitly encode multi-view parallax differences, thereby enabling integration across different LF representations. Furthermore, we introduce a Representation Aware Asymmetric Scanning (RAAS) mechanism that dynamically adjusts scanning paths based on the physical properties of different representation spaces, optimizing the balance between performance and efficiency through path pruning. Additionally, a Dual-Anchor Aggregation (DAA) module improves hierarchical feature flow, reducing redundant deeplayer features and prioritizing important reconstruction information. Experiments on various public benchmarks show that RASLF achieves the highest reconstruction accuracy while remaining highly computationally efficient.

Figures & tables

Fig. 1: Overall architecture of the proposed RASLF. (a) The Progressive Geometric Refinement (PGR) paradigm sequentially refines features using representation-specific VSSM units, and a Dual-Anchor Aggregation module fuses multi-stage features via spatial and geometric anchors. (b) Representation-Aware Asymmetric Scanning (RAAS) tailors SS2D scanning paths $\Phi$ and representation transforms $\mathcal{T},\mathcal{T}^{-1}$ for SAI, MacPI, and EPI, reducing redundant computation while preserving geometry-aware dependencies.

Fig. 2: Illustration of the Panoramic Epipolar Representation and the corresponding representation-aware scanning paths (indicated by red solid arrows).

Method	Scale	EPFL	HCINew	HCIold	INRIA	STFgantry	Average
RCAN [ 33 ]	$\times$ 2	33.156/0.9635	35.022/0.9603	41.125/0.9875	35.036/0.9769	36.670/0.9831	36.202/0.9743
resLF [ 34 ]	$\times$ 2	33.617/0.9706	36.685/0.9739	43.422/0.9932	35.395/0.9804	38.354/0.9904	37.495/0.9817
LFSSR [ 35 ]	$\times$ 2	33.671/0.9744	36.802/0.9749	43.811/0.9938	35.279/0.9832	37.944/0.9898	37.501/0.9832
LF-ATO [ 36 ]	$\times$ 2	34.272/0.9757	37.244/0.9767	44.205/0.9942	36.171/0.9842	39.636/0.9929	38.306/0.9847
MEG-Net [ 37 ]	$\times$ 2	34.312/0.9773	37.424/0.9777	44.097/0.9942	36.103/0.9849	38.767/0.9915	38.141/0.9851
DistgSSR [ 25 ]	$\times$ 2	34.809/0.9787	37.959/0.9796	44.943/0.9949	36.586/0.9859	40.404/0.9942	38.940/0.9867

TABLE I: PSNR/SSIM results compared with SOTA methods for 2

\times

and 4

\times

LFSR tasks. The best and second-best results are, respectively, in bold and underlined.

Panel of Figure 4 — Fig. 3: Qualitative visualization results for 4 $\times$ LFSR compared to other methods. Here, we showed the error maps of the reconstructed center-view images, with representative regions indicated by arrows. PSNR/SSIM values for the corresponding region are provided below.

Method	Scale	Params.	FLOPs(G)	Time(ms)	Avg. PSNR/SSIM
LFSSR [ 35 ]	$\times$ 2	0.89M	25.70	10.0	37.501/0.9832
MEG-Net [ 37 ]	$\times$ 2	1.69M	48.40	31.2	38.141/0.9851
LF-ATO [ 36 ]	$\times$ 2	1.22M	597.66	85.6	38.306/0.9847
LF-IINet [ 39 ]	$\times$ 2	5.04M	56.16	20.6	38.763/0.9860
DistgSSR [ 25 ]	$\times$ 2	3.53M	64.11	24.2	38.940/0.9867
HLFSR-SSR [ 40 ]	$\times$ 2	13.72M	167.81	31.1	39.303/0.9874

TABLE II: Comparison of parameters, flops, time, and average PSNR/SSIM values for

\times

2 and

\times

4 SR. Flops and time are calculated on an input LF with a size of

5\times 5\times 32\times 32

Fig. 4: Comparison of computational efficiency between our method and SOTA models on the

4\times

LFSR task. The area of each circle represents memory consumption.

w./o. DDA	32.637 / 0.9466	1.086	37.928
	PSNR / SIMM	Params	FLOPS
w. EPI-H & EPI-V	32.588 / 0.9466	0.953	31.200
w. EPIs-H & EPIs-V	32.628 / 0.9468	0.953	31.200
RASLF (Our)	32.690 / 0.9472	0.953	31.200

TABLE III: Ablation study on different components for 4

\times

LFSR.

	SAI	MacPI	H-EPI	V-EPI	PSNR / SIMM	Params	FLOPS
(a)	$\Phi_{4\text{-}path}$	$\Phi_{4\text{-}path}$	$\{\Phi_{row},\Phi_{row}^{{}^{\prime}}\}$	$\{\Phi_{col},\Phi_{col}^{{}^{\prime}}\}$	32.703 / 0.9473	1.013	36.050
(b)	$\{\Phi_{row},\Phi_{col}\}$	$\Phi_{4\text{-}path}$	$\{\Phi_{row},\Phi_{row}^{{}^{\prime}}\}$	$\{\Phi_{col},\Phi_{col}^{{}^{\prime}}\}$	32.700 / 0.9471	0.983	33.625
(c)	$\{\Phi_{row},\Phi_{col}\}$	$\Phi_{4\text{-}path}$	$\{\Phi_{col}\}$	$\{\Phi_{col}\}$	32.696 / 0.9471	0.953	31.200
(d)	$\{\Phi_{row},\Phi_{col}\}$	$\{\Phi_{row},\Phi_{col}\}$	$\{\Phi_{col}\}$	$\{\Phi_{col}\}$	32.329 / 0.9440	0.923	28.775

TABLE IV: Ablation study on different representation scanning strategies for 4

\times

LFSR.

M	C	PSNR / SIMM	Params	FLOPS
4	64	32.690 / 0.9472	0.953	31.200
8	64	32.901 / 0.9492	1.654	55.756
12	64	32.972 / 0.9497	2.310	76.676
16	64	33.070 / 0.9502	2.996	100.020
4	128	33.039 / 0.9498	3.486	101.185
4	256	33.129 / 0.9508	13.412	367.205

TABLE V: Investigation on the scalability of the proposed RASLF with respect to network depth (number of PGR blocks

M

) and width (feature dimensions

C

Image Super-Resolution

Image super-resolution reconstructs high-resolution images from lower-resolution observations, balancing detail, fidelity, and computational cost.

See all 248 papers

Slope-Guided Mamba and Angular-Refined Transformer for Light Field Super-Resolution
Jul 1, 2026Li Jin, Jian Huang, Junde Lu +3Image Super-Resolution
Hangzhou International Innovation Institute, Beihang University, Hangzhou, China
VSANet: View-aware Sparse Attention Network for Light Field Image Denoising
Jun 23, 2026Gargi Panda, Soumitra Kundu, Saumik Bhattacharya +1Sparse Attention Image Denoising
Department of EE, IIT Kharagpur, India · Rekhi Centre of Excellence for the Science of Happiness, IIT Kharagpur, India · Department of E&ECE, IIT Kharagpur, India
GTF: Omnidirectional EPI Transformer for Light Field Super-Resolution
May 6, 2026Kunyu Li, Fei Wang, Lichao Zhang +2Image Super-Resolution
School of Computer Science and Technology, Xi’an Shiyou University · School of Electronic Engineering, Xi’an Shiyou University · Artificial Intelligence Research Institute, Shenzhen University of Advanced Technology

Jul 1, 2026J/K move · Enter open · S save

Slope-Guided Mamba and Angular-Refined Transformer for Light Field Super-Resolution

Li Jin, Jian Huang, Junde Lu, Shuai Wang+2

Hangzhou International Innovation Institute, Beihang University, Hangzhou, China

Image Super-Resolution

Open paper AskPDF arXiv

Light Field Super-Resolution (LFSR) necessitates accurate modeling of spatial-angular correlations while preserving intrinsic 4D ray coherence. However, maintaining such high-dimensional consistency remains challenging, primarily due to two inherent limitations in prevailing modeling paradigms. First, spatial and angular dimensions are often modeled in a decoupled manner, restricting early cross-dimensional interaction and leading to geometric inconsistencies. Moreover, although continuous sequence modeling paradigms show promise in representing epipolar structures, their rigid scanning mechanisms fundamentally conflict with epipolar geometry, limiting geometry-aware feature aggregation. To address these challenges, we propose a hybrid light field super-resolution network, termed SMART, which integrates a Slope-Guided Mamba and an Angular-Refined Transformer to effectively overcome these limitations. Specifically, we introduce an angular-modulated spatial module to bridge the decoupling gap, incorporating angular priors to strengthen spatial-angular correlation modeling. To mitigate the scan-geometry mismatch, we propose a manifold-aligned trajectory module that enables geometry-consistent sequence modeling along epipolar structures. Experiments on five benchmarks demonstrate that SMART achieves state-of-the-art performance, surpassing previous methods by 0.42 dB (PSNR) with significantly reduced artifacts.

Abstract

Image Super-Resolution

Explore similar work

Slope-Guided Mamba and Angular-Refined Transformer for Light Field Super-Resolution

VSANet: View-aware Sparse Attention Network for Light Field Image Denoising

GTF: Omnidirectional EPI Transformer for Light Field Super-Resolution