Scholarlink

<p>Object detection plays a crucial role in autonomous driving. As road environments grow increasingly complex, this study proposes a novel Spatial-Channel Interaction Module to address perception challenges in autonomous driving. The module combines the local perception capabilities of convolutional neural networksss (CNNs) with the global attention mechanism of Vision Transformer (ViT). It is a lightweight, plug-and-play feature processing unit that significantly enhances object detection performance with minimal computational overhead. Experimental results demonstrate significant performance improvements in both the YOLO series and RT-DETR, particularly for the YOLOv10m model, which achieves a 2.2% increase in detection accuracy. Additionally, the precision of YOLOv10m improved by 1.3% on the MS COCO dataset. These results clearly demonstrate the adaptability and effectiveness of model in enhancing object detection, especially its potential for broad application in complex autonomous driving scenarios.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Scim: a lightweight spatial-channel interaction module for enhancing object detection in autonomous driving

  • 期刊论文

  • 发表日期：

    2026年3月12日

查看全文
  • Yang Cui,
  • Yi Han,
  • Dong Guo,
  • Tian Guan,
  • Wei Yang

摘要

Object detection plays a crucial role in autonomous driving. As road environments grow increasingly complex, this study proposes a novel Spatial-Channel Interaction Module to address perception challenges in autonomous driving. The module combines the local perception capabilities of convolutional neural networksss (CNNs) with the global attention mechanism of Vision Transformer (ViT). It is a lightweight, plug-and-play feature processing unit that significantly enhances object detection performance with minimal computational overhead. Experimental results demonstrate significant performance improvements in both the YOLO series and RT-DETR, particularly for the YOLOv10m model, which achieves a 2.2% increase in detection accuracy. Additionally, the precision of YOLOv10m improved by 1.3% on the MS COCO dataset. These results clearly demonstrate the adaptability and effectiveness of model in enhancing object detection, especially its potential for broad application in complex autonomous driving scenarios.