👋 Hello!
I am currently a Research Associate at Multimedia & Vision Group @ QMUL, working with Prof. Dimitrios Kollias on AI-Enabled Face Analysis in the Wild. In parallel, I also a Research Associate at the University of Exeter, focusing on Fairness in Personality Analysis, in collaboration with Prof. Zeyu Fu and Prof. Siyang Song.
I am a final-year Ph.D. candidate at Xi’an Jiaotong University, supervised by Prof. Xinyu Yang. From 2023 to 2024, I was a visiting researcher at Queen Mary University of London, under the supervision of Prof. Dimitrios Kollias.
🔥 I’m currently open to industry or postdoctoral opportunities starting in 2025/2026 — feel free to reach out!
🎯 Research Interests
My research focuses on Affective Analysis and Generation, encompassing Multimodal & Multitask Perception, Fair & Interpretable Modeling, and Emotion-controllable Facial Generation. The ultimate goal is to achieve human-computer interaction systems endowed with accurate, fair, and interpretable affective understanding, while exhibiting empathetic, engaging, and nuanced behaviors through full-spectrum interactions, including auditory, visual, linguistic, and action-based modalities.
Multimodel & Multitask
Multi-model Modeling
- Vision–Language Models (VLMs)
- Audio–Visual–Language Fusion
- Multimodel Prototype Alignment
- Domain Adaptation Fusion
Fair & Interpretable
Fairness & Consistency
- Fair & Consistent Analysis
- Domain-Generalized & Bias-Resilient Modeling
- Fairness-aware Multi-task optimization
Causality & Explainability
- Causal Discovery
- Counterfactual Intervention
- Human-Aligned Causality
- Visual Prototype Leaning
Featured Works: [AAAI under review], [NeurIPS under review]
Facial Generation
Talking Head Generation
- StyleGAN, Diffusion, NeRF Modeling
- Emotion-Controllable Generation
- Identity-aware Facial Motion Style Leaning and Transfer
- Multi-View & Identity-Preserving Generation
Featured Works: [Second Prize]
Reaction Generation
- Stimuli-Driven Generation
- Emotion Aware and Controllable
- Semantic-Aligned Reaction
Featured Works: [9]
✨ News
- 2025.07: 🎉 One conference paper accepted at ACM MM 2025 !
- 2025.07: 📢 Our work presented at the BMVA Symposium on AI Security !
- 2025.06: 👨💻 Served as Program Chair, Session Chair & Data Chair of the 9th ABAW Workshop and Competition at ICCV 2025 !
- 2025.04: 🎉 One journal paper accepted by IEEE Transactions on Biometrics, Behavior, and Identity Science (TBIOM) !
- 2025.01: 👨💻 Served as Program Chair, Session Chair & Data Chair of the 8th ABAW Workshop and Competition at IEEE CVPR 2025 !
- 2024.08: 🏆 Honored with the Best Reviewed Paper Award at IEEE FG 2024 !
- 2024.08: 🎉 One conference paper accepted at IEEE IJCB 2024 !
- 2024.08: 🎉 Two workshop papers accepted at ECCV 2024 [1] [2] !
- 2024.05: 👨💻 Served as Program Chair, Session Chair & Data Chair of the 7th ABAW Workshop and Competition at ECCV 2024 !
- 2025.07: 🎉 One journal paper accepted at JAMIA 2025 !
- 2024.04: 🎉 One workshop paper accepted at IEEE CVPR 2024 !
- 2024.03: 🎉 One conference paper accepted at IEEE FG 2024 !
- 2024.01: 👨💻 Served as Program Chair, Session Chair & Data Chair of the 6th ABAW Workshop and Competition at IEEE CVPR 2024 !
📝 Selected Publications
Published Databases
Towards Fair and Consistent Affect Analysis
🔗 Access Data & Code on GitHubEvaluating affective computing methods remains challenging due to inconsistencies in database partitioning and evaluation protocols, which often result in unfair or biased performance comparisons. While many studies report steady performance gains, our analysis questions the validity of such claims. Motivated by these insights, we introduce a unified evaluation protocol to ensure fairness and comparability across benchmarks.
Specifically, for six widely used affective datasets, we provide comprehensive demographic annotations—including race, gender, and age—alongside standardized evaluation metrics and a shared protocol for expression recognition, action unit detection, and valence-arousal estimation.
Additionally, we release multiple pre-trained models (ranging from baselines to state-of-the-art systems) trained under this unified protocol. A dedicated leaderboard is also introduced to encourage future research with a stronger emphasis on fairness, consistency, and reproducibility.
🏆 Honors and Awards
- 🏅 Best Reviewed Paper | IEEE FG 2024 Bridging the Gap: Protocol Towards Fair and Consistent Affect Analysis.
- 💰 Innovative Talents International Cooperative Training Project Scholarship | 2023 Fully-funded one-year overseas research visit.
- 💰 Academic Scholarship | 2023 | Xi’an Jiaotong University
- 🏆 Second Prize | 2022 | China National Radio and Television Administration Artificial Intelligence Application Innovation Competition Local Language IP Broadcasting Integration Platform Based on NeRF.
- 🏆 Excellence Award ``| 2022 | China National Radio and Television Administration New Video Technology Application Innovation Competition Intelligent Media Activation System for Video Scenarios in the Petrochemical Industry.
- 🎖️ Outstanding Student Cadre | 2022, 2023 | Xi’an Jiaotong University
- 🎖️ Outstanding Student | 2022, 2023 | Xi’an Jiaotong University
- 🎖️ Outstanding Student Scholarship | 2023 | Xi’an Jiaotong University
- 🏆 Second Prize | 2021 | China National Radio and Television Administration Artificial Intelligence Application Innovation Competition Audio-Visual Content Editing AI Platform Based on “Silk Road Cloud”.
- 🏆 Second Prize | 2021 | China National Radio and Television Administration New Video Technology Application Innovation Competition Ultra-HD Program Intelligent Production Project Based on Multimodal Fusion.
- 💰 National Encouragement Scholarship | 2016 | China Ministry of Education (MoE)