English

RT-DETR++ for UAV Object Detection

Computer Vision and Pattern Recognition 2025-09-12 v1

Abstract

Object detection in unmanned aerial vehicle (UAV) imagery presents significant challenges. Issues such as densely packed small objects, scale variations, and occlusion are commonplace. This paper introduces RT-DETR++, which enhances the encoder component of the RT-DETR model. Our improvements focus on two key aspects. First, we introduce a channel-gated attention-based upsampling/downsampling (AU/AD) mechanism. This dual-path system minimizes errors and preserves details during feature layer propagation. Second, we incorporate CSP-PAC during feature fusion. This technique employs parallel hollow convolutions to process local and contextual information within the same layer, facilitating the integration of multi-scale features. Evaluation demonstrates that our novel neck design achieves superior performance in detecting small and densely packed objects. The model maintains sufficient speed for real-time detection without increasing computational complexity. This study provides an effective approach for feature encoding design in real-time detection systems.

Keywords

Cite

@article{arxiv.2509.09157,
  title  = {RT-DETR++ for UAV Object Detection},
  author = {Yuan Shufang},
  journal= {arXiv preprint arXiv:2509.09157},
  year   = {2025}
}
R2 v1 2026-07-01T05:31:29.840Z