Scholarlink

This paper proposes a tracker that applies the Informer and sequence training method. Compared to previous Transformer-based trackers, this tracker reduces the computational power requirements and improves the speed by at least 19% while maintaining accuracy. Furthermore, we extend this tracker to the RGB-D object tracking domain, and the resulting tracker achieves a balance between accuracy and speed, enhancing the real-time operational speed to 33.2 FPS.

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

The Tracker Based on Informer

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Chunyu Chen,
  • Kaiyang Xing,
  • Yulong Qiao,
  • Wantong Sun

摘要

This paper proposes a tracker that applies the Informer and sequence training method. Compared to previous Transformer-based trackers, this tracker reduces the computational power requirements and improves the speed by at least 19% while maintaining accuracy. Furthermore, we extend this tracker to the RGB-D object tracking domain, and the resulting tracker achieves a balance between accuracy and speed, enhancing the real-time operational speed to 33.2 FPS.