Scholarlink

<p>The detection of Facial Action Units (AUs) is crucial for applications in human–computer interaction and affective computing. However, accurately identifying subtle facial muscle movements and modeling the complex relationships between different AUs remain challenging tasks. To address these challenges, we propose a novel framework comprising two key modules. First, our Region Attention Module focuses on detecting facial muscle movements without relying on any auxiliary information and achieve better AU location. Second, the AU Correlation Learning Module aims to capture intricate AU relationships by leveraging an AU label embedding method. This module represents AU-specific semantics through embeddings and encodes them with facial visual features, effectively enhancing the representation ability of AU features by incorporating learned AU relationships. Our method achieved impressive results, with scores of 65.4% on the BP4D benchmark and 65.5% on the DISFA benchmark.</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

Region attention and label embedding for facial action unit detection

  • 期刊论文

  • 发表日期：

    2025年2月18日

查看全文
  • Wei Song,
  • Dong Li

摘要

The detection of Facial Action Units (AUs) is crucial for applications in human–computer interaction and affective computing. However, accurately identifying subtle facial muscle movements and modeling the complex relationships between different AUs remain challenging tasks. To address these challenges, we propose a novel framework comprising two key modules. First, our Region Attention Module focuses on detecting facial muscle movements without relying on any auxiliary information and achieve better AU location. Second, the AU Correlation Learning Module aims to capture intricate AU relationships by leveraging an AU label embedding method. This module represents AU-specific semantics through embeddings and encodes them with facial visual features, effectively enhancing the representation ability of AU features by incorporating learned AU relationships. Our method achieved impressive results, with scores of 65.4% on the BP4D benchmark and 65.5% on the DISFA benchmark.