Short  Biography

Siyu Zhu is a full professor with tenure at College of Computer Science and Artificial Intelligence at Fudan University, where he lead the generative vision lab (Fusion Lab). He is also a full-time mentor of Shanghai Innovation Institute. Prior to this, he served as a director at Alibaba Cloud AI Lab, leading a team of 100 employees, and contributed to various aspects of computer vision and machine learning-related products. He also co-founded the 3D vision company Altizure, which was later acquired by Apple Inc and contributed core technology to Apple Augmented Reality. 

He obtained his Ph.D. from HKUST under the supervision of Long Quan, and bachelor's degree from Zhejiang University. His research interests include: (1) generative model for video and 3D; (2) diffusion-based multi-modality generative models; and (3) image based 3D reconstruction. 

He served as the area chair (senior program committee member) of ICCV 2025, CVPR 2026, ECCV 2026, NeurIPs 2026,  AAAI 2025/26/27, WACV 2025/27, publicity chair of ICCV 2025, and the associate editor of IEEE Transactions on Visualization and Computer Graphics (IEEE TVCG). He was awarded the CCF Outstanding Engineer Award.

Selected  Works (Lab Github)

• Image, Video and 3D Generative Models

• World Models and Multimodal Models

• Datasets for Generative Model

Team Members

• PhD Students

• Master Students

• Researchers (Shanghai Innovation Institute & Shanghai Academy of AI for Science)

Recent Publications (Full List)

• SlotMemory: Object-Centric KV Memory for Streaming Long-Video Generation

• GroundShot: Visually Consistent Multi-Shot Long Video Generation via Entity-Grounded Shot Scheduling

• The Thinking Pixel: Recursive Sparse Reasoning in Multimodal Diffusion Latents

• UniTacVLA: Unified Tactile Understanding and Prediction in Vision Language Action Models

• Glob3R: Global Structure-from-Motion with 3D Foundation Models

• Touch-R1: Reinforcing Touch Reasoning in MLLMs

• 3DThinkVLA: Endowing Vision-Language-Action Models with Latent 3D Priors via 3D-Thinking-Guided Co-training

• Towards Consistent Video Geometry Estimation

• Hallo-Live: Real-Time Streaming Joint Audio-Video Avatar Generation with Asynchronous Dual-Stream and Human-Centric Preference Distillation

• BARD: Bridging AutoRegressive and Diffusion Vision-Language Models Via Highly Efficient Progressive Block Merging and Stage-Wise Distillation

• Tora3: Trajectory-Guided Audio-Video Generation with Physical Coherence

• Forge4d: Feed-Forward 4D Human Reconstruction and Interpolation from Uncalibrated Sparse-View Videos

• Prompt Reinjection: Alleviating Prompt Forgetting in Multimodal Diffusion Transformers

• Towards Native Generative Model for 3D Head Avatar

• T^\star: Progressive Block Scaling for Masked Diffusion Language Models Through Trajectory Aware Reinforcement Learning

• Learning Implicit Bias in Generative Spaces for Accelerating Protein Dynamics Emulation

• LaPha: Latent Poincaré Shaping for Agentic Reinforcement Learning

• WAM-Diff: A Masked Diffusion VLA Framework with MoE and Online Reinforcement Learning for Autonomous Driving

• WAM-Flow: Parallel Coarse-to-Fine Motion Planning via Discrete Flow Matching for Autonomous Driving

• MixFlow Training: Alleviating Exposure Bias with Slowed Interpolation Mixture

• CrowdGaussian: Reconstructing High-Fidelity 3D Gaussians for Human Crowd from a Single Image

• Head-wise Adaptive Rotary Positional Encoding for Fine-Grained Image Generation

• Large Depth Completion Model from Sparse Observations

• LaTo: Landmark-tokenized Diffusion Transformer for Fine-grained Human Face Editing

• Pyramidal Patchification Flow for Visual Generation