Gaze-DETR: Top-Down Guidance Through Priority Maps for Infrared Weak-Small UAV Detection with DETR

2026-07-21

AI Summary

This work addresses the challenge of detecting faint infrared small-scale UAV targets, which are highly susceptible to clutter, noise, and occlusion, and for which conventional detection methods lack explicit spatial prioritization. To this end, we propose Gaze-DETR, the first framework to integrate a biologically inspired, top-down spatial priority mechanism into the DETR architecture. Our approach employs a priority head to generate a spatial priority map, coupled with Residual Priority-guided Feature Modulation (RPFM) and Priority-guided Anchor Query Injection (PAQI) to enhance responses in high-priority regions and guide decoder queries prior to localization. We also introduce TIR-UAV120-Gaze, the first infrared UAV detection dataset annotated with eye-tracking fixation data, and train our model via multimodal supervision. Experiments demonstrate that Gaze-DETR achieves 86.18/87.08 mAP₅₀ on TIR-UAV120-Gaze and 89.00/90.43 F1 on Anti-UAV410, validating the efficacy of learned spatial prioritization.

Abstract

Infrared small target detection (ISTD) remains challenging because tiny, low-contrast targets are easily overwhelmed by clutter, noise, or occlusion. Conventional single-frame and multi-frame detectors rely on bounding-box supervision, which specifies final target locations but offers little explicit guidance for prioritizing candidate regions or preserving weak-target evidence before localization. Task-driven visual search offers such guidance: top-down goals and visual evidence jointly form a spatial priority map that ranks candidate locations. Building on this principle, we propose Gaze-DETR, a bio-inspired detector that learns an internal priority map before localization. First, a priority head predicts a normalized priority map from image features. Second, Residual Priority-Guided Feature Modulation (RPFM) enhances high-priority responses while retaining multi-scale features. Finally, Priority-Guided Anchor Query Injection (PAQI) converts high-priority locations into decoder anchor queries. We train the priority head using three supervision schemes: box-derived Gaussian maps; real-gaze maps constructed from fixation-density maps; and transferred pseudo-gaze maps learned from gaze--box relations in paired annotations and applied to Anti-UAV410 training boxes. To support the latter two schemes, we construct TIR-UAV120-Gaze with paired detection and task-driven eye-tracking annotations. On TIR-UAV120-Gaze, Gaze-DETR achieves 85.76 mAP$_{50}$ and 88.77 F1 with box-derived supervision, and 86.18 mAP$_{50}$ and 89.00 F1 with real-gaze supervision. On Anti-UAV410, it achieves 87.06 mAP$_{50}$ and 90.90 F1 with box-derived supervision, and 87.08 mAP$_{50}$ and 90.43 F1 with transferred pseudo-gaze supervision. These results show that explicit spatial-priority learning provides pre-localization guidance complementary to bounding-box supervision across annotation settings and costs.

Problem

• Infrared Small Target Detection
• Weak-Small UAV
• Spatial Priority Map
• Top-Down Guidance
• Detection in Clutter

Innovation

• priority map
• top-down guidance
• infrared small target detection
• DETR
• gaze supervision