Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Efficient Aerial Image Detection with Variable Receptive Fields

Liu Wenbin

TL;DR

This work introduces VRF-DETR, a transformer-based detector tailored for UAV imagery that tackles small, densely occluded targets under computational constraints. It combines three novel components—MSCF for adaptive multi-scale feature fusion, GConv for parameter-efficient local-context modeling, and GMCF Bottleneck for hierarchical global-local fusion—to enable dynamic receptive-field adaptation. On VisDrone2019 and DOTA v1.0, VRF-DETR achieves a leading balance of accuracy and efficiency, notably 51.4% mAP50 on VisDrone2019 with 13.5M parameters and 31.8% mAP50-95, marking a new Pareto frontier for UAV detection. The approach demonstrates robust performance on sub-10px targets and occluded scenes without anchors or NMS, highlighting its practical impact for real-time aerial perception tasks.

Abstract

Aerial object detection using unmanned aerial vehicles (UAVs) faces critical challenges including sub-10px targets, dense occlusions, and stringent computational constraints. Existing detectors struggle to balance accuracy and efficiency due to rigid receptive fields and redundant architectures. To address these limitations, we propose Variable Receptive Field DETR (VRF-DETR), a transformer-based detector incorporating three key components: 1) Multi-Scale Context Fusion (MSCF) module that dynamically recalibrates features through adaptive spatial attention and gated multi-scale fusion, 2) Gated Convolution (GConv) layer enabling parameter-efficient local-context modeling via depthwise separable operations and dynamic gating, and 3) Gated Multi-scale Fusion (GMCF) Bottleneck that hierarchically disentangles occluded objects through cascaded global-local interactions. Experiments on VisDrone2019 demonstrate VRF-DETR achieves 51.4\% mAP\textsubscript{50} and 31.8\% mAP\textsubscript{50:95} with only 13.5M parameters. This work establishes a new efficiency-accuracy Pareto frontier for UAV-based detection tasks.

An Efficient Aerial Image Detection with Variable Receptive Fields

TL;DR

This work introduces VRF-DETR, a transformer-based detector tailored for UAV imagery that tackles small, densely occluded targets under computational constraints. It combines three novel components—MSCF for adaptive multi-scale feature fusion, GConv for parameter-efficient local-context modeling, and GMCF Bottleneck for hierarchical global-local fusion—to enable dynamic receptive-field adaptation. On VisDrone2019 and DOTA v1.0, VRF-DETR achieves a leading balance of accuracy and efficiency, notably 51.4% mAP50 on VisDrone2019 with 13.5M parameters and 31.8% mAP50-95, marking a new Pareto frontier for UAV detection. The approach demonstrates robust performance on sub-10px targets and occluded scenes without anchors or NMS, highlighting its practical impact for real-time aerial perception tasks.

Abstract

Aerial object detection using unmanned aerial vehicles (UAVs) faces critical challenges including sub-10px targets, dense occlusions, and stringent computational constraints. Existing detectors struggle to balance accuracy and efficiency due to rigid receptive fields and redundant architectures. To address these limitations, we propose Variable Receptive Field DETR (VRF-DETR), a transformer-based detector incorporating three key components: 1) Multi-Scale Context Fusion (MSCF) module that dynamically recalibrates features through adaptive spatial attention and gated multi-scale fusion, 2) Gated Convolution (GConv) layer enabling parameter-efficient local-context modeling via depthwise separable operations and dynamic gating, and 3) Gated Multi-scale Fusion (GMCF) Bottleneck that hierarchically disentangles occluded objects through cascaded global-local interactions. Experiments on VisDrone2019 demonstrate VRF-DETR achieves 51.4\% mAP\textsubscript{50} and 31.8\% mAP\textsubscript{50:95} with only 13.5M parameters. This work establishes a new efficiency-accuracy Pareto frontier for UAV-based detection tasks.

Paper Structure

This paper contains 11 sections, 3 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. MSCF Module
  4. GConv Module
  5. GMCF Bottleneck
  6. Experiments
  7. Datasets
  8. Experimental Configuration
  9. Comparison Experiments
  10. Ablation Experiments
  11. Conclusion

Figures (5)

  • Figure 1: Overall architecture of VRF-DETR. RepC3 is identical to that in RT-DETR. CA and SA representing channel and spatial attention, respectively.
  • Figure 2: The architecture of MSCF.
  • Figure 3: The architecture of GConv Module.
  • Figure 4: Difference between C2f and GMCF Bottleneck. (a) C2f. (b) GMCF.
  • Figure 5: Comparison of detection performance between VRF-DETR(the first row) and RT-DETR(the second row). Red boxes highlight incorrect detections.