FSDC-DETR: A Frequency-Spatial Domain Collaborative DETR for Small Object Detection

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Advanced, extended

Summary

FSDC-DETR, a novel Frequency-Spatial Domain Collaborative Detection Transformer, addresses the challenges of small object detection (SOD) by explicitly modeling complementary spatial and frequency representations. This framework introduces three key components: Dual-Branch Frequency-Spatial Adaptive Fusion (DBFSAF) to enhance frequency diversity and adaptively capture discriminative representations, Shunt Frequency-Spatial Feature Fusion (SFS-FF) for structure-aware cross-scale feature propagation, and Frequency-Spatial Dynamic Downsampling (FSD-Down) to preserve high-frequency responses during scale transitions. FSDC-DETR achieves strong performance, improving AP by 6.4 on VisDrone-DET2019 (to 31.1 AP) and 6.6 on AITODv2 (to 31.1 AP), with gains of 6.8 and 6.9 AP for small objects, respectively. The code is publicly available.

Key takeaway

For Machine Learning Engineers developing small object detection systems, FSDC-DETR offers a robust approach to overcome limitations in preserving fine-grained details. You should consider integrating explicit frequency-spatial domain modeling, particularly its adaptive fusion and frequency-preserving downsampling techniques, to significantly boost accuracy on challenging datasets like VisDrone-DET2019 and AITODv2. Explore the provided codebase to implement these advancements and improve your model's localization precision for tiny objects.

Key insights

FSDC-DETR enhances small object detection by explicitly integrating frequency and spatial domain modeling throughout the detection pipeline.

Principles

• High-frequency components are critical for precise small object localization.
• Explicitly modeling frequency and spatial domains improves feature representation.
• Adaptive fusion mitigates frequency aliasing and truncation in dual-branch backbones.

Method

FSDC-DETR uses DBFSAF for adaptive frequency-spatial fusion, SFS-FF for cross-scale propagation, and FSD-Down (wavelet-based grouped convolution) for frequency-preserving downsampling.

In practice

• Consider frequency-spatial modeling for fine-grained object detection tasks.
• Explore wavelet decomposition for frequency-aware downsampling in multi-scale networks.
• Utilize the FSDC-DETR codebase for SOD research or applications.

Topics

• Small Object Detection
• Detection Transformers
• Frequency Domain Analysis
• Spatial-Frequency Fusion
• Deep Learning Architectures
• Wavelet Transforms

Code references

• nevereverinsomnia/FSDC-DETR

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• FSDETR: Frequency-Spatial Feature Enhancement for Small Object Detection — FSDETR enhances small object detection by integrating frequency and spatial features to overcome common detection challenges.
• Demystifying RF-DETR [ICLR 2026]: A Real-Time Transformer Pushing the Limits of Object Detection — RF-DETR uses weight-sharing NAS and internet-scale pre-training to achieve state-of-the-art real-time object detection and segmentation.
• A Spatial-Spectral-Frequency Interactive Network for Multimodal Remote Sensing Classification — Integrating spatial, spectral, and frequency domain fusion enhances multimodal remote sensing classification, especially with limited labels.
• ViCrop-Det: Spatial Attention Entropy Guided Cropping for Training-Free Small-Object Detection — ViCrop-Det improves small-object detection by adaptively cropping regions using spatial attention entropy, without retraining.
• Towards Multi-Object-Tracking with Radar on a Fast Moving Vehicle: On the Potential of Processing Radar in the Frequency Domain — Frequency domain processing of radar data enhances robustness and inherently detects all moving objects for multi-object tracking.
• Real-Time Source-Free Object Detection — RT-SFOD achieves state-of-the-art source-free object detection by optimizing pseudo-label fusion and multi-scale feature diversification for speed and accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.