Multi-View Single-Scan Visual State-Space Network for Efficient Image Super-Resolution

Area of Science:

• Computer Vision
• Deep Learning
• Image Processing

Background:

• Single Image Super-Resolution (SISR) is crucial for edge devices with limited resources.
• Existing visual state space models struggle with 2D dependencies, increasing computation and memory usage.

Purpose of the Study:

• To develop an efficient SISR network for edge devices.
• To address the computational and memory limitations of current models.

Main Methods:

• Propose the Multi-View Single-Scan Visual State Space Network (MVSSN).
• Utilize a shuffled input stacking module (SISM) for efficient feature extraction.
• Employ a multi-view single-scan block (MSSB) for lightweight cross-layer context modeling.
• Incorporate a multi-scale local feature block (MLFB) for detailed local feature restoration.

Main Results:

• MVSSN achieves competitive or superior PSNR and SSIM metrics on standard SISR benchmarks.
• The network operates with fewer than one million parameters and low FLOPs.
• Evaluations demonstrate efficiency and practical behavior in RealSR and downstream tasks.

Conclusions:

• MVSSN offers an efficient solution for SISR on edge devices.
• The proposed architecture effectively balances performance, parameter count, and computational cost.
• Model limitations under unknown real-world degradation warrant further investigation.