More than a Feeling: The MiFace Framework for Defining Facial Communication Mappings
summary
MiFace presents a reproducible pipeline that generates FACS-based avatar expressions using Faceshift morphs (28 AUs) and crowdsourced MTurk judgments, filters 1,609 computationally generated images down to 220 reliably perceived expressions, and produces 51 unique labels via GloVe similarity and hierarchical clustering (cutoff 0.8375). Two proof‑of‑concept studies (three‑AU and two‑AU combos) test believability then single‑word labeling; outcome claims an ~8x expansion of labeled expression space and demonstrates the approach works at scale while noting limits in avatar realism and synonym scoring.
AI-written summary · about
chapters (12)
- 0:00Introduction to MiFace
- 0:31What is facial communication mapping
- 3:53Limitations of current expression mappings
- 6:24Facial expressions in technology
- 8:37Introducing the MiFace framework
- 9:42Building the FACS-based avatar
- 10:45Proof of concept studies setup
- 11:50Crowd-sourced judgment studies
- 13:19Label clustering and analysis
- 15:48Results and expression mappings
- 16:29Future work and applications
- 17:11Q&A session
related talks
10mNon-isomorphic Interaction Techniques for Controlling Avatar Facial Expressions in VR
07mFaraPy: An Augmented Reality Feedback System for Facial Paralysis using Action Unit Intensity ...
36mEmbodied AI: A Cognitive Shift, Not a Cosmetic Change
16mRepresentation in the Metaverse: Building avatars & identity for 3B people
32mDesigning Beyond the Screen
32mUnderstanding and Measuring User Emotions: The Missing Puzzle Piece
09mGazeChat: Enhancing Virtual Conferences with Gaze Awareness and Interactive 3D Photos
37mPersonality Is the New Ringtone: Experiments in AI design