Yiyuan Zhang | 张懿元

PhD Candidate @ CUHK MMLabBytedance Seed

I am currently a third-year Ph.D. candidate at the MMLab, CUHK. I work on pretraining and modeling at Bytedance Seed, where I was fortunate to develop frontier foundation models, including the Seed 2.0, Seed 1.8, and Seed1.5-VL.

Driven by a philosophy of Simplicity, Scalability, and Systematization, my research bridges architectural innovation and large-scale pretraining to build end-to-end systems with real-world capabilities.

Email Google Scholar Github
Yiyuan Zhang | 张懿元

Job market 2027. Looking for opportunities (industry/postdoc). Please feel free to reach out!

News

  • [03/2026]Two papers accepted to CVPR 2026!
  • [02/2026]We present Seed 2.0 Pro, which ranks 3rd on the Vision Arena.
  • [07/2025]Four papers accepted to ICCV 2025!
  • [06/2025]Journal extension of UniRepLKNet is accepted to IEEE TPAMI.
  • [02/2024]Four papers accepted to CVPR 2024.
  • [10/2023]Meta-Transformer was officially reported by MIT TechReview.

Work Experience

ByteDance, Seed-VL Team

  • Core contributors, in Doubao-1.5-Pro-Vision, Seed1.5-VL, Seed-1.8, and Seed-2.0.
  • Foundation model team of VLM @ Bytedance Seed.
  • Working on VLM architecture, pretraining, video understanding, and unified understanding & generation.
Seed-1.5-VL

Seed-1.5-VL

Core Contributors

Foundation VLM model focusing on advanced visual understanding.

Seed-2.0 & Seed-1.8

Seed-2.0 & Seed-1.8

Core Contributors

Towards Intelligence Frontier for Real-World Complexity

Education

The Chinese University of Hong Kong

PhD in Department of Information Engineering (MMLab)
08/2023 - 06/2027 (Expected)
Hong Kong

Beijing Institute of Technology

Bachelor in Artificial Intelligence
36th XuTeLi Scholarship (Top 10/8000+)
09/2019 - 06/2023
Beijing

Research Publications

Categorized by theme. Full list on Google Scholar.

Multimodal Scalability
Learning Beyond Still Frames: Scaling Vision-Language Models with Video

Learning Beyond Still Frames: Scaling Vision-Language Models with Video

Yiyuan Zhang, Handong Li, Jing Liu, Xiangyu Yue

ICCV 2025. IEEE
Explores the scaling laws of integrating continuous video data into existing VLM architectures.
Paper
Scaling Omni-modal Pretraining with Multimodal Context: Advancing Universal Representation Learning Across Modalities

Scaling Omni-modal Pretraining with Multimodal Context: Advancing Universal Representation Learning Across Modalities

Yiyuan Zhang, Handong Li, Jing Liu, Xiangyu Yue

ICCV 2025. IEEE
First scalable omnimodal pretraining scheme. Officially reported by MIT TechReview.
Paper Code
Scaling Up Your Kernels: Large Kernel Design in ConvNets towards Universal Representations

Scaling Up Your Kernels: Large Kernel Design in ConvNets towards Universal Representations

Yiyuan Zhang, Xiaohan Ding, Xiangyu Yue

IEEE Transactions on Pattern Analysis and Machine Intelligence 2025.
Comprehensive analysis and structural design principles for ultra-large kernels in modern ConvNets.
Paper GitHub
Multimodal Simplicity
[1] Meta-Transformer: A Unified Framework for Multimodal Learning

[1] Meta-Transformer: A Unified Framework for Multimodal Learning

Yiyuan Zhang, Kaixiong Gong, Kaipeng Zhang, Hongsheng Li, Wanli Ouyang, Yu Qiao, Xiangyu Yue

Arxiv Preprint
We found a unified framework performing multimodal learning across 12 modalities. Featured by MIT TechReview.
Paper GitHub Project
[2] UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio Video Point Cloud Time-Series and Image Recognition

[2] UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio Video Point Cloud Time-Series and Image Recognition

Xiaohan Ding*, Yiyuan Zhang*, Yixiao Ge, Sijie Zhao, Lin Song, Xiangyu Yue, Ying Shan

CVPR 2024. IEEE
We further explore the universal modeling capacity of Transformers and propose a universal perception large-kernel convnet for multimodal recognition.
Paper GitHub Project
[3] OneLLM: One Framework to Align All Modalities with Language

[3] OneLLM: One Framework to Align All Modalities with Language

Jiaming Han, Kaixiong Gong, Yiyuan Zhang, Jiaqi Wang, Kaipeng Zhang, Dahua Lin, Yu Qiao, Peng Gao, Xiangyu Yue

CVPR 2024. IEEE
We propose an elegant framework to align diverse modalities seamlessly with LLMs.
Paper Project GitHub
[4] OneThinker: All-in-one Reasoning Model for Image and Video

[4] OneThinker: All-in-one Reasoning Model for Image and Video

Kaituo Feng, Manyuan Zhang, Hongyu Li, Kaixuan Fan, Shuang Chen, Yilei Jiang, Dian Zheng, Peiwen Sun, Yiyuan Zhang, Haoze Sun, Yan Feng, Peng Pei, Xunliang Cai, Xiangyu Yue

CVPR 2026. IEEE
We propose a unified reasoning model for image and video understanding, which achieves state-of-the-art performance on various vision-language reasoning tasks.
Paper Code
Multimodal System
Multimodal Pathway: Improve Transformers with Irrelevant Data from Other Modalities

Multimodal Pathway: Improve Transformers with Irrelevant Data from Other Modalities

Yiyuan Zhang, Xiaohan Ding, Kaixiong Gong, Yixiao Ge, Ying Shan, Xiangyu Yue

CVPR 2024. IEEE
Demonstrates how irrelevant data across modalities can structurally improve transformer representations.
Paper Project Code
Asymmetric Convolution: An Efficient and Generalized Method to Fuse Feature Maps in Multiple Vision Tasks

Asymmetric Convolution: An Efficient and Generalized Method to Fuse Feature Maps in Multiple Vision Tasks

Wencheng Han, Xingping Dong, Yiyuan Zhang, David Crandall, Cheng-Zhong Xu, Jianbing Shen

IEEE Transactions on Pattern Analysis and Machine Intelligence 2024.
A generalized feature fusion method accelerating multiple down-stream vision tasks.
Paper
Seed1.5-VL Technical Report

Seed1.5-VL Technical Report

Bytedance Seed Team

Technical Report
Technical report of Seed1.5-VL, which ranks 1st in Chinese Vision Arena.
Paper Project
Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity

Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity

Bytedance Seed Team

Model Card
Model card of Seed2.0, which ranks 3rd in the international Vision Arena.
Paper Project

Open-Source Impact

Simple Scaling-S1

Main Contributor

⭐ 6.6k

Meta-Transformer

Projector Owner

⭐ 1.7k

UniRepLKNet

Projector Owner

⭐ 1.1k

OneLLM

Main Contributor

⭐ 665

Professional Service

  • Conference Reviewer: CVPR, NeurIPS, ICML, ICLR, ICCV, ECCV, AAAI, etc.
  • Journal Reviewer: IEEE TPAMI, TIP, TNNLS, IJCV, etc.
  • Invited Talks: OpenMMlab, TechBeat, etc.