FlowIR: Depth-driven Rectified Flow for Resolving Thermal Ambiguity in Infrared Super-Resolution
Authors: Xingyue Zhu, Qinyun Mei, Zhexi Min, Yang Zou, Xingyuan Li, Zhiying Jiang, Jinyuan Liu
Organizations: Dalian University of Technology, Dalian, China · Northwestern Polytechnical University, Xi’an, China · Dalian Maritime University, Dalian, China
Distinguishing different heat sources in low-resolution infrared (IR) images is inherently difficult, as multiple thermal emitters often collapse into ambiguous and overlapping responses. Existing super-resolution methods hallucinate structures from such ambiguous 2D evidence, leading to thermal drift, boundary merging, and unreliable restoration under infrared degradations. We propose FlowIR, a depth-driven rectified flow framework that resolves thermal ambiguity by injecting 3D geometric cues into the generation process. Pseudo-3D depth lifting and depth-guided 3D convolutions disentangle overlapping foreground objects, while cluster-wise inter-attention suppresses cross-source feature contamination. A Thermal-Geometry Consistency Loss further enforces geometric alignment during refinement. Built upon a lightweight rectified flow backbone, FlowIR restores IR images in only a few steps, producing sharply separated, degradation-robust, and geometrically faithful high-resolution results.
Figures & tables
Fig. 1: (a) Overlapping heat sources introduce thermal ambiguity in SR. (b) Prior methods rely solely on 2D features. (c) Grayscale fluctuations show that FlowIR better preserves pixel fidelity in overlapping heat sources regions. (d) FlowIR mitigates thermal ambiguity by integrating 3D geometric guidance into a rectified flow framework.
Fig. 2: Overview of FlowIR. a) The overall pipeline, a depth-driven rectified flow framework that integrates depth-derived 3D cues into rectified flow to produce structurally faithful and degradation-robust reconstructions. b) The Depth-Aware Point Cloud Condition (DAPCC) module, which provides explicit geometric cues to mitigate overlapping heat sources by injecting depth-aware guidance into the flow trajectory. c) The 3D Cluster-Guided Decoder (3DCD) that enforces feature consistency within each semantic cluster while preventing high-frequency leakage across different objects.
Datasets
Set5
Set10
Set20
Methods
PSNR ↑
SSIM ↑
LPIPS ↓
CLIP-IQA ↑
MUSIQ ↑
PSNR ↑
SSIM ↑
LPIPS ↓
CLIP-IQA ↑
MUSIQ ↑
PSNR ↑
SSIM ↑
LPIPS ↓
CLIP-IQA ↑
MUSIQ ↑
FMB
ChasNet
42.2995
0.9732
0.1587
0.1432
27.9576
42.0907
0.9429
0.1607
0.1436
28.3499
42.0137
0.9713
0.1612
0.1445
28.1572
DAT
42.4312
0.9740
0.1587
0.1462
30.3660
42.3480
0.9713
0.1566
0.1435
30.0319
42.4958
0.9777
0.1560
0.1508
30.3948
DifIISR
33.3211
0.9073
0.1780
0.3414
46.2157
33.0764
0.9462
0.1812
0.3463
46.4277
34.4098
0.9461
0.1761
0.3381
45.9119
HAT
42.4071
0.9741
0.1590
0.1490
29.6160
42.7815
0.9677
0.1569
0.1489
29.9870
42.0661
0.9428
0.1582
0.1527
29.9735
TABLE I: Quantitative comparison on FMB and M3FD ( × 4). The best is in bold , while the second is underlined . For FMB, Set5/10/20 are randomly sampled subsets. For M3FD , the three sets correspond to three degradation levels of increasing severity.
Fig. 3: Qualitative results. The first and second rows are from the static scenes of the FMB dataset ( × 4), while the third and fourth rows are from the M3FD dataset ( × 4).
Fig. 4: Quantitative ablation study on Depth Awareness.
Fig. 5: Qualitative ablation study on Depth Awareness.
Fig. 6: Quantitative ablation study of individual modules on PSNR , SSIM , MUSIQ .
Fig. 7: Qualitative ablation study on individual modules.
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.
Axi Niu, Zhenguo Wu, Kang Zhang +3
School of Computer Science, Northwestern Polytechnical University, Xi’an, Shaanxi, China · School of Electrical Engineering, KAIST, Daejeon, Republic of Korea · School of Aeronautics and Astronautics, Northwestern Polytechnical University, Xi’an, Shaanxi, China
Infrared image super-resolution (IISR) under real-world conditions is a practically significant yet rarely addressed task. Pioneering works are often trained and evaluated on simulated datasets or neglect the intrinsic differences between infrared and visible imaging. In practice, however, real infrared images are affected by coupled optical and sensing degradations that jointly deteriorate both structural sharpness and thermal fidelity. To address these challenges, we propose Real-IISR, a unified autoregressive framework for real-world IISR that progressively reconstructs fine-grained thermal structures and clear backgrounds in a scale-by-scale manner via thermal-structural guided visual autoregression. Specifically, a Thermal-Structural Guidance module encodes thermal priors to mitigate the mismatch between thermal radiation and structural edges. Since non-uniform degradations typically induce quantization bias, Real-IISR adopts a Condition-Adaptive Codebook that dynamically modulates discrete representations based on degradation-aware thermal priors. Also, a Thermal Order Consistency Loss enforces a monotonic relation between temperature and pixel intensity, ensuring relative brightness order rather than absolute values to maintain physical consistency under spatial misalignment and thermal drift. We build FLIR-IISR, a real-world IISR dataset with paired LR-HR infrared images acquired via automated focus variation and motion-induced blur. Extensive experiments demonstrate the promising performance of Real-IISR, providing a unified foundation for real-world IISR and benchmarking. The dataset and code are available at: https://github.com/JZD151/Real-IISR.
Yang Zou, Jun Ma, Zhidong Jiao +3
School of Computer Science, Northwestern Polytechnical University · School of Software Technology & DUT-RU International School of ISE, Dalian University of Technology · School of Computer Science, Zhejiang University +1
Remote sensing infrared image super-resolution aims to recover sharper thermal observations from low-resolution inputs while preserving target contours, scene layout, and radiometric stability. Unlike visible-image super-resolution, thermal imagery is weakly textured and more sensitive to unstable local sharpening, which makes complementary local and global modeling especially important. This paper presents our solution to the NTIRE 2026 Infrared Image Super-Resolution Challenge, a dual-branch system that combines a HAT-L branch and a MambaIRv2-L branch. The inference pipeline applies test-time local conversion on HAT, eight-way self-ensemble on MambaIRv2, and fixed equal-weight image-space fusion. We report both the official challenge score and a reproducible evaluation on 12 synthetic times-four thermal samples derived from Caltech Aerial RGB-Thermal, on which the fused output outperforms either single branch in PSNR, SSIM, and the overall Score. The results suggest that infrared super-resolution benefits from explicit complementarity between locally strong transformer restoration and globally stable state-space modeling.
Xining Ge, Gengjia Chang, Weijun Yuan +6
Hangzhou Dianzi University · Hefei University of Technology · Jinan University +1