Scholarlink

Deploying YOLO11 on resource-constrained platforms is crucial for real-time ADAS applications. This paper explores structured pruning, unstructured pruning, and quantization to optimize YOLO11 for deployment on Raspberry Pi 5 and Jetson AGX Orin. We evaluate model variants using mean Average Precision (mAP) and power consumption. Structured pruning improves efficiency with minimal accuracy loss, while unstructured pruning better preserves accuracy. Quantization, especially Post-training Quantization (PTQ), achieves significant power and memory savings. Results highlight the need to tailor compression strategies to hardware for effective ADAS deployment.

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

Optimized YOLO11 Deployment on ECU for ADAS: Balancing Accuracy and Power with Model Compression

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Anirudh Nayak,
  • P. C. Nissimagoudar,
  • H. M. Gireesha,
  • K. H. Aarya,
  • Jyoti Ravikumar,
  • Nalini C. Iyer

摘要

Deploying YOLO11 on resource-constrained platforms is crucial for real-time ADAS applications. This paper explores structured pruning, unstructured pruning, and quantization to optimize YOLO11 for deployment on Raspberry Pi 5 and Jetson AGX Orin. We evaluate model variants using mean Average Precision (mAP) and power consumption. Structured pruning improves efficiency with minimal accuracy loss, while unstructured pruning better preserves accuracy. Quantization, especially Post-training Quantization (PTQ), achieves significant power and memory savings. Results highlight the need to tailor compression strategies to hardware for effective ADAS deployment.