Research map / Graph learning / Knowledge and temporal graphs
ReactioNet: Learning High-Order Facial Behavior from Universal Stimulus-Reaction by Dyadic Relation Reasoning
Xiaotian Li, Taoyue Wang, Geran Zhao, Xiang Zhang, Xi Kang, Lijun Yin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add cross-domain contrastive learning (CDC) (ablation) | DRR · ReactioNet AUD | F1: 58.9 -> 61.3 | Improved |
| add homophily relation module (HO) (ablation) | DRR · ReactioNet AUD | F1: 59.1 -> 61.3 | Improved |
| add heterophily relation module (HE) (ablation) | DRR · ReactioNet scene classification | accuracy: 56.1 -> 60.2 | Improved |
| add heterophily relation module (HE) (ablation) | DRR · ReactioNet facial expression recognition | accuracy: 73.5 -> 78.5 | Improved |
| add stimulus and text domain information (DRR dynamic vs ResNet18) | ResNet-18 · ReactioNet AUD | F1: 54.8 -> 61.3 | Improved |
| feature-fusion of reaction and stimulus without relation reasoning | Feature fusion baseline · ReactioNet AUD | F1: 54.8 -> 54.6 | No clear effect |
| use temporal sequential frames (DRR dynamic) instead of single frame (DRR static) | DRR · ReactioNet facial expression recognition | accuracy: 76.2 -> 78.5 | Improved |
| multi-task dyadic reasoning (DRR dyadic) instead of dynamic on AUD | DRR · ReactioNet AUD | F1: 61.3 -> 61.1 | No clear effect |
| apply DRR static to film and sports scenes | DRR Static · ReactioNet AUD by scene | F1: ResNet-18 53.9 -> 50.4 on film scene | Hurt |
| pretrain on ReactioNet (DRR PR) and fine-tune on benchmark datasets | visual encoder · RAFDB FER | accuracy: 79.1 -> 89.6 (vs baseline GCN) | Improved |
| DRR variants on ten benchmark datasets | DRR · 5 AUD and 5 FER datasets | F1 or accuracy: best on seven of ten | Mixed |
| train on ReactioNet and test on other datasets versus the reverse | ResNet-18 · cross-database BP4D, DISFA, EmotioNet, AffectNet | F1 and accuracy: up to 23% | Improved |
Closest papers by meaning
- Facial-R1: Aligning Reasoning and Recognition for Facial Emotion Analysis (AAAI 2026 · similarity 0.88)
- PerReactor: Offline Personalised Multiple Appropriate Facial Reaction Generation (AAAI 2025 · similarity 0.87)
- OUS: Bridging Scene Context and Facial Features to Overcome the Rigid Cognitive Problem (AAAI 2025 · similarity 0.86)
- AFFAKT: A Hierarchical Optimal Transport Based Method for Affective Facial Knowledge Transfer in Video Deception Detection (AAAI 2025 · similarity 0.86)
- GLARE: Generating Listening Heads with Appropriate REactions (NeurIPS 2026 · similarity 0.86)
- EmoNet-Face: An Expert-Annotated Benchmark for Synthetic Emotion Recognition (NeurIPS 2025 · similarity 0.86)
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.
