FR-DETR: Frequency and Recurrent Feature Refinement for Robust Object Detection under Adverse Weather

2026-06-29

AI Summary

This work addresses the significant performance degradation of object detectors under adverse weather conditions, primarily caused by visual degradation and domain shift. To tackle this issue, the authors propose a detector-centric feature optimization framework that dynamically fuses frequency- and spatial-domain information at the feature level. Specifically, a frequency refinement module adaptively reweights frequency components within regions of interest, while a recursive focus refinement module (RFRM), guided by coarse predictions, iteratively enhances discriminative features. By operating directly on features rather than performing image-level enhancement, the method avoids redundant computation and focuses explicitly on improving detection performance. Experimental results demonstrate that the proposed approach substantially improves detection accuracy across various adverse weather conditions while significantly reducing computational overhead, outperforming existing enhancer-based methods.

Abstract

Object detection under adverse weather remains challenging due to severe visual degradations and domain shifts. Existing enhancer-based approaches attempt to improve detection by cascading an enhancer with a detector, but they introduce redundant feature extraction and incur high computational cost with limited accuracy gains when paired with SOTA detectors. We propose FR-DETR, a detector-centric framework that refines features rather than images, focusing enhancement on regions of interest and leveraging frequency-domain cues. Specifically, we design (I) a Frequency Refinement Module that dynamically separates and reweights low- and high-frequency components to improve foreground-background discrimination, and (II) a Recurrent Focus Refinement Module (RFRM) that iteratively refines features using coarse predictions as guidance. Extensive experiments demonstrate that FR-DETR achieves superior detection accuracy under adverse weather while being significantly more computationally efficient than enhancer-based methods. Our implementation is available at https://github.com/ducnt1210/FR-DETR.

Problem

• adverse weather
• object detection
• visual degradation
• domain shift

Innovation

• Frequency Refinement
• Recurrent Feature Refinement
• Detector-Centric Framework
• Adverse Weather Object Detection
• Feature-Level Enhancement