Research map / 3D vision / Face and shape
Biomechanics-Guided Facial Action Unit Detection Through Force Modeling
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| combine 3D force and position instead of using either alone (ablation) | 3D physics branch · BP4D | average F1 (%): 46.4 and 53.9 -> 57.3 | Helped |
| combine the physics branch with the image branch (ablation) | AU detection model · BP4D | average F1 (%): 57.3 and 58.7 -> 64.1 | Helped |
| add 3D reconstruction loss (ablation) | AU detection model · BP4D, 50% training data | average F1 (%): 39.0 -> 46.2 | Helped |
| add 3D reconstruction loss (ablation) | AU detection model · BP4D, 20% training data | average F1 (%): 34.7 -> 44.3 | Helped |
| add 3D reconstruction loss (ablation) | AU detection model · BP4D, 5% training data | average F1 (%): 27.9 -> 33.0 | Helped |
| use the biomechanics-based physics branch rather than the data-driven model | cross-dataset AU detection · BP4D to DISFA (overlapping AUs) | average F1 (%): 32.1 -> 38.8 | Helped |
| use the proposed method instead of FAU | AU detection · BP4D | average F1 (%): 64.2 -> 64.1 | Hurt |
| use the proposed method instead of FAU | AU detection · DISFA | average F1 (%): 61.5 -> 58.2 | Hurt |
| use biomechanics-based force prediction instead of image-only prediction (ablation) | AU1 detection · BP4D | F1 (%): 46.3 -> 58.6 | Helped |
| use biomechanics-based force prediction instead of image-only prediction (ablation) | AU15 detection · BP4D | F1 (%): 48.1 -> 29.4 | Hurt |
| add force and position physics features to image features (ablation) | AU detection model · BP4D | F1 (%): AU1: 46.3 -> 57.4; AU15: 48.1 -> 47.3 | Mixed |
Closest papers by meaning
- Electromyography-Informed Facial Expression Reconstruction for Physiological-Based Synthesis and Analysis (CVPR 2025 · similarity 0.84)
- Muscles in Time: Learning to Understand Human Motion In-Depth by Simulating Muscle Activations (NeurIPS 2024 · similarity 0.84)
- PhysPT: Physics-aware Pretrained Transformer for Estimating Human Dynamics from Monocular Videos (CVPR 2024 · similarity 0.83)
- AUEditNet: Dual-Branch Facial Action Unit Intensity Manipulation with Implicit Disentanglement (CVPR 2024 · similarity 0.82)
- DIFFTACTILE: A Physics-based Differentiable Tactile Simulator for Contact-rich Robotic Manipulation (ICLR 2024 · similarity 0.81)
- DeformMaster: An Interactive Physics-Neural World Model for Deformable Objects from Videos (NeurIPS 2026 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.