Scholarlink

<p>Industrial big data visualization poses significant challenges for diverse user groups with varying cognitive levels and background knowledge, affecting their ability to interpret results swiftly and accurately. Unlike traditional methods that rely on eye-tracking devices, this study presents GazeMT, an innovative hybrid converter model that enables accurate gaze estimation without the need for traditional eye-tracking hardware. We introduce gaze estimation model with multi-scale features and transformer(GazeMT), a hybrid Transformer model that integrates multi-scale feature aggregation and a shared attention mechanism for precise gaze estimation. GazeMT utilizes ResNet with a CBAM (convolutional block attention module) to extract multi-scale features, which are then processed by a Transformer to capture global dependencies. An improved shared attention mechanism further optimizes gaze estimation by leveraging sample similarities, enhancing fine-grained detail capture. Extensive experiments on public datasets (MPIIFaceGaze, RT-Gene) show that the average angular error of GazeMT is reduced by 0.07° and 0.82° on MPIIFaceGaze and RT-Gene, respectively. Visualizing gaze points as scatter plots provides an intuitive view of user attention distribution, offering a scientific basis for optimizing industrial big data visualization design. This research has important implications for personalized interface design and enhanced user experience. The code address of this paper is: <a href="https://github.com/IRlexx/GazeMT">https://github.com/IRlexx/GazeMT</a>.</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

Enhancing industrial big data visualization through multi-scale gaze estimation and personalized dashboard design

  • 期刊论文

  • 发表日期：

    2025年7月9日

查看全文
  • Liang Chen,
  • Huangguo Peng

摘要

Industrial big data visualization poses significant challenges for diverse user groups with varying cognitive levels and background knowledge, affecting their ability to interpret results swiftly and accurately. Unlike traditional methods that rely on eye-tracking devices, this study presents GazeMT, an innovative hybrid converter model that enables accurate gaze estimation without the need for traditional eye-tracking hardware. We introduce gaze estimation model with multi-scale features and transformer(GazeMT), a hybrid Transformer model that integrates multi-scale feature aggregation and a shared attention mechanism for precise gaze estimation. GazeMT utilizes ResNet with a CBAM (convolutional block attention module) to extract multi-scale features, which are then processed by a Transformer to capture global dependencies. An improved shared attention mechanism further optimizes gaze estimation by leveraging sample similarities, enhancing fine-grained detail capture. Extensive experiments on public datasets (MPIIFaceGaze, RT-Gene) show that the average angular error of GazeMT is reduced by 0.07° and 0.82° on MPIIFaceGaze and RT-Gene, respectively. Visualizing gaze points as scatter plots provides an intuitive view of user attention distribution, offering a scientific basis for optimizing industrial big data visualization design. This research has important implications for personalized interface design and enhanced user experience. The code address of this paper is: https://github.com/IRlexx/GazeMT.