What question did this study set out to answer?

To develop SSD-SR, a super-resolution architecture leveraging Mamba-2's Structured State Space Duality for improved image and video quality.

February 21, 2026Open Access

SSD-SR: Upgrading Image and Video Super-Resolution from Mamba-1 to Mamba-2 with Structured State Space Duality

Key Points

To develop SSD-SR, a super-resolution architecture leveraging Mamba-2's Structured State Space Duality for improved image and video quality.
Introduced SSD Residual Groups for enhanced processing
Implemented Frequency-Aware State Routing for better frequency handling
Developed Diffusion-Guided SSD-SR as a diffusion-based approach to super-resolution
Demonstrated improved image and video quality over previous Mamba-1 methods
Achieved effective differentiation between low and high-frequency signals
Establishes a new benchmark for super-resolution in Mamba frameworks

Abstract

We propose SSD-SR, the first super-resolution architecture built on Mamba-2's Structured State Space Duality (SSD). While over 15 Mamba-1 SR methods exist (MambaIR, MambaIRv2, DVMSR, VSRM, etc.), none uses Mamba-2. SSD-SR introduces SSD Residual Groups, Frequency-Aware State Routing for differential low/high-frequency processing, and Diffusion-Guided SSD-SR — the first diffusion-based Mamba SR model.

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper