Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

D$^3$R-DETR: DETR with Dual-Domain Density Refinement for Tiny Object Detection in Aerial Images

Zixiao Wen, Zhen Yang, Xianjie Bao, Lei Zhang, Xiantai Xiang, Wenshuai Li, Yuhan Liu

TL;DR

D3R-DETR tackles tiny object detection in aerial imagery by introducing Dual-Domain Density Refinement to a DETR-based framework. It fuses spatial context through a DilatedSPU and frequency-domain features via Fractional Gabor Kernels in a Dual-Domain Fusion Module (D2FM), complemented by a lightweight density head supervised with Density Recall Focal Loss to generate accurate density maps that guide query generation. Evaluations on AI-TOD-v2 show state-of-the-art performance and faster convergence, particularly in high-density tiny-object scenes, validating the effectiveness of dual-domain guidance for query–object matching. The approach enhances detection robustness in remote sensing and offers a path toward integrating temporal and semantic information for further gains.

Abstract

Detecting tiny objects plays a vital role in remote sensing intelligent interpretation, as these objects often carry critical information for downstream applications. However, due to the extremely limited pixel information and significant variations in object density, mainstream Transformer-based detectors often suffer from slow convergence and inaccurate query-object matching. To address these challenges, we propose D$^3$R-DETR, a novel DETR-based detector with Dual-Domain Density Refinement. By fusing spatial and frequency domain information, our method refines low-level feature maps and utilizes their rich details to predict more accurate object density map, thereby guiding the model to precisely localize tiny objects. Extensive experiments on the AI-TOD-v2 dataset demonstrate that D$^3$R-DETR outperforms existing state-of-the-art detectors for tiny object detection.

D$^3$R-DETR: DETR with Dual-Domain Density Refinement for Tiny Object Detection in Aerial Images

TL;DR

D3R-DETR tackles tiny object detection in aerial imagery by introducing Dual-Domain Density Refinement to a DETR-based framework. It fuses spatial context through a DilatedSPU and frequency-domain features via Fractional Gabor Kernels in a Dual-Domain Fusion Module (D2FM), complemented by a lightweight density head supervised with Density Recall Focal Loss to generate accurate density maps that guide query generation. Evaluations on AI-TOD-v2 show state-of-the-art performance and faster convergence, particularly in high-density tiny-object scenes, validating the effectiveness of dual-domain guidance for query–object matching. The approach enhances detection robustness in remote sensing and offers a path toward integrating temporal and semantic information for further gains.

Abstract

Detecting tiny objects plays a vital role in remote sensing intelligent interpretation, as these objects often carry critical information for downstream applications. However, due to the extremely limited pixel information and significant variations in object density, mainstream Transformer-based detectors often suffer from slow convergence and inaccurate query-object matching. To address these challenges, we propose DR-DETR, a novel DETR-based detector with Dual-Domain Density Refinement. By fusing spatial and frequency domain information, our method refines low-level feature maps and utilizes their rich details to predict more accurate object density map, thereby guiding the model to precisely localize tiny objects. Extensive experiments on the AI-TOD-v2 dataset demonstrate that DR-DETR outperforms existing state-of-the-art detectors for tiny object detection.
Paper Structure (11 sections, 2 equations, 5 figures, 2 tables)

This paper contains 11 sections, 2 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overview
  4. Dual-Domain Density Refinement
  5. Dual-Domain Fusion Module
  6. Lightweight Density Head
  7. Results
  8. Dataset and Implementation Details
  9. Comparison with state-of-the-art
  10. Ablation Study
  11. Discussion

Figures (5)

  • Figure 1: Overview architecture of our proposed D3R-DETR. D2FM fuses spatial and frequency domain information to extract richer features for accurate density map reconstruction, along with a lightweight density head. MWAS denotes Masked Window Attention Sparsification, and PAQI denotes Progressive Adaptive Query Initialization---both adopted from Dome-DETR hu2025dome. CCFF denotes CNN-based Cross-scale Feature Fusion zhao2024detrs.
  • Figure 2: Visualization of FrGK in different angles and scales.
  • Figure 3: The proposed DCBlock in DilatedSPU.
  • Figure 4: AP Performance and DRFL Comparisons.
  • Figure 5: Qualitative results in AI-TOD-v2 test dataset. Top row: results of the baseline model; Bottom row: results of D3R-DETR. The green, red, and blue boxes represent TP, FP, and FN, respectively.