YOLO-DS: Fine-Grained Feature Decoupling via Dual-Statistic Synergy Operator for Object Detection

TL;DR

YOLO-DS introduces a novel feature decoupling framework with dual-statistic modeling, significantly improving object detection accuracy across multiple scales while maintaining efficiency.

Contribution

The paper proposes YOLO-DS with a Dual-Statistic Synergy Operator and lightweight gating modules, enhancing feature discrimination in YOLO detectors for better performance.

Findings

YOLO-DS outperforms YOLOv8 on MS-COCO with 1.1%-1.7% AP gains.

The approach maintains minimal inference latency increase.

Extensive validation confirms improved heterogeneous object discrimination.

Abstract

One-stage object detection, particularly the YOLO series, strikes a favorable balance between accuracy and efficiency. However, existing YOLO detectors lack explicit modeling of heterogeneous object responses within shared feature channels, which limits further performance gains. To address this, we propose YOLO-DS, a framework built around a novel Dual-Statistic Synergy Operator (DSO). The DSO decouples object features by jointly modeling the channel-wise mean and the peak-to-mean difference. Building upon the DSO, we design two lightweight gating modules: the Dual-Statistic Synergy Gating (DSG) module for adaptive channel-wise feature selection, and the Multi-Path Segmented Gating (MSG) module for depth-wise feature weighting. On the MS-COCO benchmark, YOLO-DS consistently outperforms YOLOv8 across five model scales (N, S, M, L, X), achieving AP gains of 1.1% to 1.7% with only a minimal increase in inference latency. Extensive visualization, ablation, and comparative studies validate the effectiveness of our approach, demonstrating its superior capability in discriminating heterogeneous objects with high efficiency.