Scholarlink

<p>To initiate machine learning (ML)-based drift chamber track reconstruction, for which no dedicated dataset currently exists, we present DCTracks-v1.0—a Monte Carlo (MC) dataset comprising single- and two-track events, with the latter including both conventional and close-by topologies. To enable standardized evaluation, we adopt a set of track reconstruction evaluation metrics, with which we benchmark both traditional and graph neural network (GNN)-based methods on this dataset. Our results show that while the GNN-based approach performs comparably to traditional methods on single and conventional two-track events, its performance degrades on close-by two-track events, highlighting both the potential and current limitations of GNN-based tracking. By providing baseline results on this dataset, this work validates the reliability of the dataset and establishes a foundation for systematic, reproducible validation and fair comparison in future research.</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

DCTracks: a dataset for machine learning-based drift chamber track reconstruction

  • 期刊论文

  • 发表日期：

    2026年6月25日

查看全文
  • Liyan Qian,
  • Yao Zhang,
  • Ye Yuan,
  • Zhaoke Zhang,
  • Jin Fang,
  • Shimiao Jiang,
  • Jin Zhang,
  • Ke Li,
  • Beijiang Liu,
  • Chenglin Xu,
  • Yifan Zhang,
  • Xiaoqian Jia,
  • Xiaoshuai Qin,
  • Xingtao Huang

摘要

To initiate machine learning (ML)-based drift chamber track reconstruction, for which no dedicated dataset currently exists, we present DCTracks-v1.0—a Monte Carlo (MC) dataset comprising single- and two-track events, with the latter including both conventional and close-by topologies. To enable standardized evaluation, we adopt a set of track reconstruction evaluation metrics, with which we benchmark both traditional and graph neural network (GNN)-based methods on this dataset. Our results show that while the GNN-based approach performs comparably to traditional methods on single and conventional two-track events, its performance degrades on close-by two-track events, highlighting both the potential and current limitations of GNN-based tracking. By providing baseline results on this dataset, this work validates the reliability of the dataset and establishes a foundation for systematic, reproducible validation and fair comparison in future research.