A novel Multi to Single Module for small object detection

TL;DR

This paper introduces the Multi to Single Module (M2S), a novel approach that enhances small object detection by improving feature extraction and refinement, leading to better accuracy on drone-based datasets.

Contribution

The paper proposes the M2S module with CAM and DRM to improve small object detection, a novel design that enhances feature extraction and adds an extra detection head.

Findings

M2S improves detection accuracy by 1.1% on VisDrone2021-DET.

M2S achieves a 15.68% accuracy increase on SeaDronesSeeV2.

Experimental results outperform existing methods.

Abstract

Small object detection presents a significant challenge in computer vision and object detection. The performance of small object detectors is often compromised by a lack of pixels and less significant features. This issue stems from information misalignment caused by variations in feature scale and information loss during feature processing. In response to this challenge, this paper proposes a novel the Multi to Single Module (M2S), which enhances a specific layer through improving feature extraction and refining features. Specifically, M2S includes the proposed Cross-scale Aggregation Module (CAM) and explored Dual Relationship Module (DRM) to improve information extraction capabilities and feature refinement effects. Moreover, this paper enhances the accuracy of small object detection by utilizing M2S to generate an additional detection head. The effectiveness of the proposed method is evaluated on two datasets, VisDrone2021-DET and SeaDronesSeeV2. The experimental results demonstrate its improved performance compared with existing methods. Compared to the baseline model (YOLOv5s), M2S improves the accuracy by about 1.1\% on the VisDrone2021-DET testing dataset and 15.68\% on the SeaDronesSeeV2 validation set.