MambaX: Image Super-Resolution with State Predictive Control

2025-11-22

AI Summary

Existing image super-resolution methods struggle to effectively suppress error propagation and accumulation across intermediate reconstruction stages. Although Mamba-style approaches support state sequence modeling, their fixed linear projectors exhibit narrow receptive fields and limited flexibility, hindering fine-grained reconstruction. To address this, we propose the Dynamic State Prediction and Control (DSPC) framework, which introduces nonlinear spectral-band-to-hidden-state sequence modeling, learnable parameterization of control equations, and a state cross-regulation paradigm for adaptive intervention on intermediate reconstruction states. Furthermore, we design a progressive transition learning strategy to mitigate heterogeneity across modalities and domains. Extensive experiments demonstrate that DSPC achieves state-of-the-art performance on both single-image and multimodal super-resolution benchmarks, with significant improvements in cross-spectral and cross-dimensional generalization capabilities.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Intelligent Robots: State Estimation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Image super-resolution (SR) is a critical technology for overcoming the inherent hardware limitations of sensors. However, existing approaches mainly focus on directly enhancing the final resolution, often neglecting effective control over error propagation and accumulation during intermediate stages. Recently, Mamba has emerged as a promising approach that can represent the entire reconstruction process as a state sequence with multiple nodes, allowing for intermediate intervention. Nonetheless, its fixed linear mapper is limited by a narrow receptive field and restricted flexibility, which hampers its effectiveness in fine-grained images. To address this, we created a nonlinear state predictive control model extbf{MambaX} that maps consecutive spectral bands into a latent state space and generalizes the SR task by dynamically learning the nonlinear state parameters of control equations. Compared to existing sequence models, MambaX 1) employs dynamic state predictive control learning to approximate the nonlinear differential coefficients of state-space models; 2) introduces a novel state cross-control paradigm for multimodal SR fusion; and 3) utilizes progressive transitional learning to mitigate heterogeneity caused by domain and modality shifts. Our evaluation demonstrates the superior performance of the dynamic spectrum-state representation model in both single-image SR and multimodal fusion-based SR tasks, highlighting its substantial potential to advance spectrally generalized modeling across arbitrary dimensions and modalities.

Problem

• Overcoming hardware limitations in image super-resolution sensors
• Addressing error propagation in intermediate reconstruction stages
• Improving fine-grained image reconstruction with nonlinear state control

Innovation

• Dynamic state predictive control learning nonlinear coefficients
• State cross-control paradigm for multimodal fusion
• Progressive transitional learning mitigates domain heterogeneity