Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hierarchical Space-Time Attention for Micro-Expression Recognition

Haihong Hao, Shuo Wang, Huixia Ben, Yanbin Hao, Yansong Wang, Weiwei Wang

TL;DR

We address micro-expression recognition by introducing Hierarchical Space-Time Attention (HSTA), which integrates Unimodal Space-Time Attention (USTA) and Crossmodal Space-Time Attention (CSTA) in a hierarchical framework to model temporal dynamics and crossmodal cues from video frames and special frames. USTA captures temporal relations within each modality, while CSTA fuses modalities with preserved modality-specific content, and the stacked HSTA structure learns deeper facial cues. Across four benchmarks, HSTA achieves state-of-the-art results, notably yielding significant gains on CASME3 while maintaining computational efficiency far below large pre-trained models like Micro-BERT. The method demonstrates strong generalization, robust performance with respect to data modalities, and valuable insights from ablations on temporal frame counts and hierarchical configuration.

Abstract

Micro-expression recognition (MER) aims to recognize the short and subtle facial movements from the Micro-expression (ME) video clips, which reveal real emotions. Recent MER methods mostly only utilize special frames from ME video clips or extract optical flow from these special frames. However, they neglect the relationship between movements and space-time, while facial cues are hidden within these relationships. To solve this issue, we propose the Hierarchical Space-Time Attention (HSTA). Specifically, we first process ME video frames and special frames or data parallelly by our cascaded Unimodal Space-Time Attention (USTA) to establish connections between subtle facial movements and specific facial areas. Then, we design Crossmodal Space-Time Attention (CSTA) to achieve a higher-quality fusion for crossmodal data. Finally, we hierarchically integrate USTA and CSTA to grasp the deeper facial cues. Our model emphasizes temporal modeling without neglecting the processing of special data, and it fuses the contents in different modalities while maintaining their respective uniqueness. Extensive experiments on the four benchmarks show the effectiveness of our proposed HSTA. Specifically, compared with the latest method on the CASME3 dataset, it achieves about 3% score improvement in seven-category classification.

Hierarchical Space-Time Attention for Micro-Expression Recognition

TL;DR

We address micro-expression recognition by introducing Hierarchical Space-Time Attention (HSTA), which integrates Unimodal Space-Time Attention (USTA) and Crossmodal Space-Time Attention (CSTA) in a hierarchical framework to model temporal dynamics and crossmodal cues from video frames and special frames. USTA captures temporal relations within each modality, while CSTA fuses modalities with preserved modality-specific content, and the stacked HSTA structure learns deeper facial cues. Across four benchmarks, HSTA achieves state-of-the-art results, notably yielding significant gains on CASME3 while maintaining computational efficiency far below large pre-trained models like Micro-BERT. The method demonstrates strong generalization, robust performance with respect to data modalities, and valuable insights from ablations on temporal frame counts and hierarchical configuration.

Abstract

Micro-expression recognition (MER) aims to recognize the short and subtle facial movements from the Micro-expression (ME) video clips, which reveal real emotions. Recent MER methods mostly only utilize special frames from ME video clips or extract optical flow from these special frames. However, they neglect the relationship between movements and space-time, while facial cues are hidden within these relationships. To solve this issue, we propose the Hierarchical Space-Time Attention (HSTA). Specifically, we first process ME video frames and special frames or data parallelly by our cascaded Unimodal Space-Time Attention (USTA) to establish connections between subtle facial movements and specific facial areas. Then, we design Crossmodal Space-Time Attention (CSTA) to achieve a higher-quality fusion for crossmodal data. Finally, we hierarchically integrate USTA and CSTA to grasp the deeper facial cues. Our model emphasizes temporal modeling without neglecting the processing of special data, and it fuses the contents in different modalities while maintaining their respective uniqueness. Extensive experiments on the four benchmarks show the effectiveness of our proposed HSTA. Specifically, compared with the latest method on the CASME3 dataset, it achieves about 3% score improvement in seven-category classification.
Paper Structure (21 sections, 17 equations, 3 figures, 7 tables)

This paper contains 21 sections, 17 equations, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Micro-expression Recognition
  4. Attention Calculation
  5. Approach
  6. Preliminaries
  7. Unimodal Space-Time Attention
  8. Crossmodal Space-Time Attention
  9. Hierarchical Learning
  10. Experiments
  11. Experimental Settings
  12. Dataset
  13. Evaluation Metrics
  14. Implementation Details
  15. Ablation Studies
  16. ...and 6 more sections

Figures (3)

  • Figure 1: The overview of our Hierarchical Space-Time Attention (HSTA). Our model is a hierarchical structure that captures the underlying emotions in micro-expression videos through multiple attention modules.
  • Figure 2: The details of USTA. (a) The cascaded structure of USTA. (b) The calculations of $l^{\rm th}$ layer USTA.
  • Figure 3: Performances of frames with different numbers in USTA.