FaithIR: Rethinking Infrared Image Super-Resolution from Perceptual Sharpness to Task Relevant Fidelity

2026-08-04

AI Summary

This work addresses the limitations of existing infrared image super-resolution methods, which often introduce artificial textures and spurious high-frequency details that compromise genuine thermal structures and semantic content, thereby degrading downstream machine perception tasks. To mitigate this issue, the authors propose FaithIR, a framework that prioritizes structural fidelity for machine perception over mere visual sharpness. FaithIR employs a dual-branch architecture: a block-level conditional branch captures global thermal structure to provide structural priors, while a structure-guided pixel-level reconstruction branch refines local details, with both branches trained end-to-end entirely in the pixel domain. Experiments demonstrate that FaithIR achieves superior reconstruction fidelity and strong generalization across the FLIR-IISR, M3FD, and FMB datasets, significantly enhancing performance in object detection and semantic segmentation.

Abstract

Infrared image super-resolution (IISR) is important for downstream tasks such as object detection and semantic segmentation. Existing IISR methods often produce artificial textures, over-sharpened edges, and spurious high-frequency details that distort authentic thermal structures and semantic information. To address this issue, we propose FaithIR, a faithful infrared super-resolution framework for reliable machine perception. FaithIR consists of a patch-level conditioning branch that captures global thermal and structural information and a pixel-level restoration branch that performs dense local reconstruction under structural guidance. The entire restoration process is performed directly in the pixel domain to preserve infrared-specific structures and task-relevant information. Extensive experiments on FLIR-IISR, M3FD, and FMB demonstrate strong reconstruction fidelity, cross-dataset generalization, and superior performance in object detection and semantic segmentation. These results show that demonstrate that preserving faithful infrared structure preservations is more important for reliable machine perception than merely pursuing perceptual sharpness alone.

Problem

• Infrared image super-resolution
• artificial textures
• structural distortion
• task-relevant fidelity
• machine perception

Innovation

• infrared image super-resolution
• task-relevant fidelity
• dual-branch architecture
• pixel-domain restoration
• machine perception