Inter-Stance: A Dyadic Multimodal Corpus for Conversational Stance Analysis
Paper Overview
- Field: Computer Vision (CV)
- Authors: Xiang Zhang, Xiaotian Li, Taoyue Wang
- Released: 2025-04-28
- arXiv: 2504.19769
- Scale: 45 dyads / 90 participants; ~20TB of multimodal data
- Modalities: 2D face video, 3D face geometry, thermal dynamics, voice and speech, physiology (PPG, EDA, heart rate, blood pressure, respiration), self-reported affect
- Dyad types: acquaintances with shared history and strangers
- Stance labels: agreement, disagreement, neutral
- Availability: planned to be shared with the research community
Abstract
Social interactions dominate our perceptions of the world and shape our daily behavior by attaching social meaning to acts as simple as gestures, facial expressions, voice, and speech. The authors present a new data corpus of multimodal dyadic interaction (45 dyads, 90 persons) that includes synchronized multi-modality behavior: 2D face video, 3D face geometry, thermal spectrum dynamics, voice and speech behavior, physiology (PPG, EDA, heart rate, blood pressure, and respiration), and self-reported affect of all participants.
Two types of dyads are included: persons with a shared past history and strangers. Annotations include social signals, agreement, disagreement, and neutral stance. With a potent emotion induction protocol, these multimodal data are intended to enable novel modeling of multimodal interpersonal behavior.