Fei Shen

Fei Shen

Sharing🌱, Collaborating🤝, Advancing🚀, Innovating💡

About Me

I am a Research Fellow at the NExT++ Center, National University of Singapore (NUS), working with Prof. Tat-Seng Chua. Previously, I received my Ph.D. from Nanjing University of Science and Technology under Prof. Jinhui Tang, and conducted research at Tencent AI Lab with Prof. Xiao Han and Dr. Jun Zhang. I was very lucky to participate in several interesting projects, including Duix.Avatar Duix.Avatar stars, V-Express V-Express stars, IMAGHarmony IMAGDHarmony stars, and IMAGDressing IMAGDressing stars. To date, I have achieved 50+ top-three finishes in algorithm competitions, including five CCF-A workshop championships, and maintain close collaborations with Tencent AI Lab, Huawei, AIUNI, GuijiAI, and Mobvoi. I serve as a reviewer for TPAMI, NeurIPS, ICML, ICLR, CVPR, ECCV, and ICCV. PS: We are actively seeking motivated Ph.D., M.S., and undergraduate students to collaborate with Prof. Chua and myself through CSC, RA, visiting positions, or remote internships in the following research areas. If you are interested, please send your CV to shenfei29@nus.edu.sg.

📚 Research Interests

My research focuses on uncovering and utilizing latent knowledge in large foundation models (LFMs), with an emphasis on multilingual, multimodal, and multi-agent systems. At a conceptual level, this line of work is inspired by the philosophical view that diverse observable behaviors may arise from deeper shared structures: from Plato’s notion of abstract Forms to Aristotle’s distinction between potentiality and actuality. I study how such latent knowledge is structured, how it gives rise to core capabilities (e.g., understanding, reasoning, and safety), and how these capabilities can be activated, aligned, protected, and transferred across models and domains. In parallel, I explore controllable image and video generation to enable reliable and fine-grained manipulation of visual content.

Research Directions

  • Latent Knowledge of LFMs
    • Specific and Shared Representations Analysis: Understanding how latent knowledge is structured across language-, modality-, and agent-specific as well as shared representations, and how a unified semantic space emerges across multilingual, multimodal, and multi-agent systems.
    • Alignment, Transfer, and Enhancement: Studying how latent capabilities can be activated, aligned, protected, and transferred across languages, modalities, agents, and model architectures, enabling scalable and robust capability reuse.
    • Representative Capability Domains: Understanding and reasoning; safety and alignment; cultural awareness; forgery and anomaly detection.
  • Image/Video Generation and Editing
    • Human- and Story-Centric Generation: Generating coherent human-centric and story-driven visual content.
    • Fashion and Garment Generation: Modeling fine-grained appearance and structure for clothing and fashion applications.
    • Controllable Editing: Enabling precise and interpretable manipulation of visual content.
    • Modular and Adapter-based Generation: Developing plug-and-play and scalable generation frameworks.

🔥 News

  • 2026.04:  🎉 Three papers were accepted by ACL 2026 (2 Main & 1 Findings).
  • 2026.03:  🎉 One paper was accepted by IEEE TVCG.
  • 2026.02:  🎉 Two papers were accepted by CVPR 2026 (1 Highlight).
  • 2026.01:  🎉 One paper was accepted by IEEE TIP.
  • 2025.11:  🎉 Seven papers were accepted by AAAI 2026, including 5 Orals.
  • 2025.09:  🎉 Three papers were accepted by NeurIPS 2025.
  • 2025.05:  🎉 Two papers were accepted by ICML 2025.
  • 2025.01:  🎉 Two papers were accepted by ICLR 2025.
  • 2024.12:  🎉 Five papers were accepted by AAAI 2025.
  • 2024.10:  🎉 One paper was accepted by ACM ToMM.
  • 2024.09:  🎉 One paper was accepted by NeurIPS 2024.
  • 2024.01:  🎉 One paper was accepted by ICLR 2024.
  • 2023.07:  🎉 One paper was accepted by ACM Multimedia 2023.


📝 Selected Publications

latent-anomaly-knowledge-excavation
Latent Anomaly Knowledge Excavation: Unveiling Sparse Sensitive Neurons in Vision-Language Models
S. Li, S. Li, C. Shi, W. Wu, Y. Wu, X. Yu, F. Shen✉, T.-S. Chua
Under Review.
[PDF]    [Code]
beyond-surface-artifacts
Beyond Surface Artifacts: Capturing Shared Latent Forgery Knowledge Across Modalities
J. Dou, C. Shi, J. Wang, F. Shen✉, Z. Wang, T.-S. Chua
Under Review.
[PDF]    [Code]
versavogue
VersaVogue: Visual Expert Orchestration and Preference Alignment for Unified Fashion Synthesis
J. Yu, F. Shen†, C. Wang, Y. Xin, S. Shen, X. Du, J. Tang
Under Review.
[PDF]    [Code]
transport-and-merge
Transport and Merge: Cross-Architecture Merging for Large Language Models
C. Cui, B. Yang, F. Shen✉, Y. Chen, J. Zheng, X. Wang, A. Zhang, T.-S. Chua
Under Review.
[PDF]    [Code]
who-transfers-safety
Who Transfers Safety? Identifying and Targeting Cross-Lingual Shared Safety Neurons
X. Zhang, C. Xie, L. Zhu, Y. Yang, W. Zhao, Z. Cheng, C. Wang,
F. Shen✉, T.-S. Chua
Under Review.
[PDF]    [Code]
dna
DNA: Uncovering Universal Latent Forgery Knowledge
J. Dou, C. Shi, Y. Wang, S. Guo, A. Yi, W. Wu, L. Zhang,
F. Shen✉, T.-S. Chua
Under Review.
[PDF]    [Code]
lingua-safetybench
Lingua-SafetyBench: A Benchmark for Safety Evaluation of Multilingual Vision-Language Models
E. Shi, P. Shao, Y. Zhang, C. Cui, J. Lyu, X. Xie, X. Xia, F. Shen✉, T.-S. Chua
Under Review.
[PDF]    [Code]
orthoeraser
OrthoEraser: Coupled-Neuron Orthogonal Projection for Concept Erasure
C. Shi, W. Wu, F. Shen✉, X. Zhu, K. Hu, Z. Wang
Under Review.
[PDF]    [Code]
tracerouter
TraceRouter: Robust Safety for Large Foundation Models via Path-Level Intervention
C. Shi, S. Li, W. Lu, W. Wu, C. Wang, Z. Cheng, F. Shen✉, T.-S. Chua
Under Review.
[PDF]    [Code]
where-culture-fades
Where Culture Fades: Revealing the Cultural Gap in Text-to-Image Generation
C. Shi, S. Li, S. Guo, S. Xie, W. Wu, J. Dou, C. Wu, C. Xiao, C. Wang, Z. Cheng, F. Shen✉, T.-S. Chua
The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR (Highlight)), 2026.
[PDF]    [Code]
imagagent
IMAGAgent: Orchestrating Multi-Turn Image Editing via Constraint-Aware Planning and Reflection
F. Shen, C. Xie, L. Wang, Z. Zhang, X. Jiang, X. Du, J. Tang
Under Review.
[PDF]    [Code]
imagedit
IMAGEdit : Let Any Subject Transform
F. Shen, W. Xu, R. Yan, D. Zhang, X. Shu, J. Tang
Under Review.
[PDF]    [Code]
imagharmony
IMAGHarmony: Controllable image editing with consistent object quantity and layout
F. Shen, X. Du, Y. Gao, J. Yu, Y. Cao, X. Lei, J. Tang
Under Review.
[PDF]    [Code]
imaggarmentplus
IMAGGarment+: Efficient Attribute-Wise Diffusion for Garment Generation
J. Yu*, F. Shen*, C. Wang, Y. Sun, H. Tang, Q. Guo, X. Du
AAAI Conference on Artificial Intelligence (AAAI), 2026.
[PDF]    [Code]
imaggarment
IMAGGarment: Fine-grained garment generation for controllable fashion design
F. Shen, J. Yu, C. Wang, X. Jiang, X. Du, J. Tang
IEEE Transactions on Visualization and Computer Graphics, (IEEE TVCG), 2026.
[PDF]    [Code]
sgmhand
SGMHand: Structure-Guided Modulation for Structure-Aware Hand Inpainting
C. Shi, S. Guo, Y. Chen, K. Shui, F. Shen✉
AAAI Conference on Artificial Intelligence (AAAI), 2026.
[PDF]    [Code]
CoC-VLA
CoC-VLA: Delving into Adversarial Domain Transfer for Explainable Autonomous Driving via Chain-of-Causality Visual-Language-Action Model
D. Zhang, F. Shen✉, R. Zhao, Y. Chen, P. Zhi, C. Li, R. Zhou, Q. Zhou
Neural Information Processing Systems (NeurIPS), 2025.
[PDF]    [Code]
IMAGDressing-v1
IMAGDressing-v1: Customizable Virtual Dressing
F. Shen, X. Jiang, X. He, H. Ye, C. Wang, X. Du, Z. Li, J. Tang
AAAI Conference on Artificial Intelligence (AAAI), 2025.
[PDF]    [Code]
MCDM
Long-Term TalkingFace Generation via Motion-Prior Conditional Diffusion Model
F. Shen, C. Wang, J. Gao, Q. Guo, J. Dang, J. Tang, T.-S. Chua
International Conference on Machine Learning (ICML), 2025.
[PDF]    [Code]
RCDMs
Boosting Consistency in Story Visualization with Rich-Contextual Conditional Diffusion Models
F. Shen, H. Ye, S. Liu, J. Zhang, C. Wang, X. Han, W. Yang
AAAI Conference on Artificial Intelligence (AAAI), 2025.
[PDF]    [Code]
IMAGPose
IMAGPose: A Unified Conditional Framework for Pose-Guided Person Generation
F. Shen, J. Tang
Neural Information Processing Systems (NeurIPS), 2024.
[PDF]    [Code]
PCDMs
Advancing Pose-Guided Image Synthesis with Progressive Conditional Diffusion Models
F. Shen, H. Ye, J. Zhang, C. Wang, X. Han, W. Yang
International Conference on Learning Representations (ICLR), 2024.
[PDF]    [Code]
triplet-contrastive-vehreid
Triplet Contrastive Representation Learning for Unsupervised Vehicle Re-Identification
F. Shen, X. Du, L. Zhang, X. Shu, J. Tang
ACM Transactions on Multimedia Computing, Communications and Applications (ToMM), 2024
.
[PDF]    [Code]
GiT: Graph Interactive Transformer for Vehicle Re-identification
GiT: Graph Interactive Transformer for Vehicle Re-identification
F. Shen, Y. Xie, J. Zhu, X. Zhu, H. Zeng
IEEE Transactions on Image Processing (IEEE TIP), 2023
[PDF]    [Code]
hybrid-pyramidal-graph-network
Exploring Spatial Significance via Hybrid Pyramidal Graph Network for Vehicle Re-Identification
F. Shen, J. Zhu, X. Zhu, Y. Xie, J. Huang
IEEE Transactions on Intelligent Transportation Systems, (IEEE TITS), 2023.
[PDF]    [Code]
PBSL: Pedestrian-specific Bipartite-aware Similarity Learning for Text-based Person Retrieval
Pedestrian-specific Bipartite-aware Similarity Learning for Text-based Person Retrieval
F. Shen, X. Shu, X. Du, J. Tang
ACM International Conference on Multimedia (ACM MM), 2023.
[PDF]    [Code]