Unified Multimodal Autoregressive Modeling with Shared ContextβVisual Tokenizer is Key to Unification
ICML, 2026
Wujian Peng ε½θι΄
Ph.D. Candidate
School of Computer Science,
Fudan University & Shanghai Innovation Institute
Biography
I am a fourth-year Ph.D. candidate at Fudan University and Shanghai Innovation Institute , supervised by Prof. Zuxuan Wu. Before that, I received my B.S. degree from Xi'an Jiaotong University in 2022. I am currently working as a research intern at Qwen Team , supervised by Shuai Bai and Lingchen Meng.
My current research interests include:
- Large Multimodal Models: Enhancing fine-grained recognition and understanding capabilities of large multimodal models, enabling more accurate and detailed comprehension of the real world.
- Unified Multimodal Models: Building unified architectures to jointly model multimodal tasks such as generation and understanding.
I am expected to graduate in 2027 and am actively seeking job opportunities. Please feel free to reach out to me via email: wjpeng22[AT]m[DOT]fudan[DOT]edu[DOT]cn .
Publications
* denotes equal contribution, † denotes corresponding author
GitHub
Compositional Text-to-Image Generation Via Region-aware Bimodal Direct Preference Optimization
CVPR, 2026
INST-IT: Boosting Instance Understanding via Explicit Visual Prompt Instruction Tuning
NeurIPS, 2025
Synthesize, Diagnose, and Optimize: Towards Fine-Grained Vision-Language Understanding
CVPR, 2024
BMB: Balanced Memory Bank for Long-Tailed Semi-Supervised Learning
TMM, 2024
Building an Open-Vocabulary Video CLIP Model with Better Architectures, Optimization and Data
TPAMI, 2024
Comp: Continual Multimodal Pre-training for Vision Foundation Models
Under review
Honors & Awards
- Outstanding Student, Fudan University
- Outstanding Student, Xi'an Jiaotong University
Professional Service
- CVPR 2025β2026 Reviewer
- ICCV 2025 Reviewer
- ICML 2024β2026 Reviewer
- NeurIPS 2024β2025 Reviewer