AuxDet: Auxiliary Metadata Matters for Omni-Domain Infrared Small Target Detection

TL;DR

Omni-domain infrared small target detection faces significant cross-domain shifts across platforms, spectral bands, and resolutions. The authors introduce AuxDet, a metadata-guided multimodal detector that dynamically fuses auxiliary metadata with visual features through a Multi-Modal Dynamic Modulation (M2DM) module and refines edges with a Lightweight Edge Enhancement Module (LEEM). On the WideIRSTD-Full benchmark, AuxDet achieves state-of-the-art performance in $AP_{50}$ and Recall and remains effective when paired with Transformer backbones, demonstrating strong cross-domain generalization and robustness. The work underscores auxiliary metadata as a practical, low-cost prior to enhance omni-domain IRSTD performance and suggests broad applicability to multimodal perception tasks in challenging imaging conditions.

Abstract

Omni-domain infrared small target detection (Omni-IRSTD) poses formidable challenges, as a single model must seamlessly adapt to diverse imaging systems, varying resolutions, and multiple spectral bands simultaneously. Current approaches predominantly rely on visual-only modeling paradigms that not only struggle with complex background interference and inherently scarce target features, but also exhibit limited generalization capabilities across complex omni-scene environments where significant domain shifts and appearance variations occur. In this work, we reveal a critical oversight in existing paradigms: the neglect of readily available auxiliary metadata describing imaging parameters and acquisition conditions, such as spectral bands, sensor platforms, resolution, and observation perspectives. To address this limitation, we propose the Auxiliary Metadata Driven Infrared Small Target Detector (AuxDet), a novel multimodal framework that is the first to incorporate metadata into the IRSTD paradigm for scene-aware optimization. Through a high-dimensional fusion module based on multi-layer perceptrons (MLPs), AuxDet dynamically integrates metadata semantics with visual features, guiding adaptive representation learning for each individual sample. Additionally, we design a lightweight prior-initialized enhancement module using 1D convolutional blocks to further refine fused features and recover fine-grained target cues. Extensive experiments on the challenging WideIRSTD-Full benchmark demonstrate that AuxDet consistently outperforms state-of-the-art methods, validating the critical role of auxiliary information in improving robustness and accuracy in omni-domain IRSTD tasks. Code is available at https://github.com/GrokCV/AuxDet.

Figures (7)

• Figure 1: Traditional vs. unified detection. AuxDet leverages metadata for effective omni-domain generalization using one unified model.
• Figure 2: The benefit of leveraging auxiliary information to assist infrared small target detection lies in its ability to guide the network towards inference objectives that typically require deep architectures, but at an extremely low computational cost.
• Figure 3: Multidimensional distribution characteristics of the WideIRSTD-Full Dataset. (a) Cross-platform target-background contrast: Resolution-normalized visualization of platform-specific heterogeneity in background complexity and small target morphology. (b) Cross-platform resolution distribution: x/y-axes represent log-transformed image height and width ranges, revealing resolution disparities across observation platforms (Air/Land/Space). (c) Platform-band correlation chord diagram: Coupling relationships between observation platforms (Air/Land/Space) and infrared bands (Long-Wave Infrared, LWIR; Near-Infrared, NIR; Short-Wave Infrared, SWIR). Key findings include: (1) Inter-platform resolution intersection: Significant overlapping regions between Air (green) and Land (blue) platforms necessitate cross-platform multi-scale detection; (2) Platform-dependent target characteristics: Distinct platform-specific distributions of target resolutions and categories induce varying detection difficulties; (3) Spectral-platform association rules: Space-based observations (orange) are NIR-dominant, while aerial/terrestrial platforms (green/blue) solely utilize LWIR.
• Figure 4: Overall architecture of AuxDet. Infrared images are processed through a backbone network to extract hierarchical features, while auxiliary metadata undergo pre-encoding and fusion via MLPs. The two modalities enter into M2DM module to achieve dynamic feature interaction and instance-level spatial-channel modulation, obtaining modulated features that are subsequently refined by LEEM. The optimized features are fused through the Feature Pyramid Network (FPN) and fed into the detection head for target localization.
• Figure 5: Architecture of the M2DM module.