I am currently a joint Ph.D. candidate of Shanghai Jiao Tong University and Monash University from SEIEE since 2022, working with Prof. Weiyao Lin and Prof. Mehrtash Harandi. Before that, I spent 4 wonderful years studying at Huazhong University of Science and Technology as an undergraduate student since 2018 (top 5%). My research interests lie in visual generation, visual understanding, and noise learning, with a particular focus on Diffusion Transformer (DiT). Currently, my work centers on Unified Visual Generation and Understanding with DiT, aiming to overcome the long-standing disconnect between these two tasks and enable them to mutually enhance each other.

If you are interested in my work, please contact me via ganchaofan@sjtu.edu.cn.

🎯 Research Interests

• Diffusion Transformers: Visual Generation and Representation Learning with DiTs, et al.
• Image/Video Understanding: Multimodal Large Models, et al.
• Noise Learning: Self-supervised Semantics Discovery, et al.

🔥 News

• 2025.9:  🎉🎉 1 first-author NeurIPS is accepted!
• 2024.10:  ✈️✈️ I will attend ACMMM conference in Melbourne in Oct. Looking forward to seeing old/new friends!
• 2024.9:  🎉🎉 1 NeurIPS(Spotlight) is accepted!
• 2024.7:  🎉🎉 1 first-author ACMMM is accepted!

📝 Selected Projects

📦 Diffusion Transformers

ARXIV 2025, preprint

Massive Activations are the Key to Local Detail Synthesis in Diffusion Transformers

Chaofan Gan,Zicheng Zhao, Yuanpeng Tu, Xi Chen, Ziran Qin, Tieyuan Chen, Yuxi Li, Mehrtash Harandi, Weiyao Lin

ARXIV 2025, preprint

Arxiv|Project|Github

• Massive Activations Drive Fine-Grained Local Detail Synthesis in DiTs.
• Detail Guidance: Superior Local Detail Fidelity Compared to CFG.

NeurIPS 2025

Unleashing Diffusion Transformers for Visual Correspondence by Modulating Massive Activations

Chaofan Gan, Yuanpeng Tu, Xi Chen, Tieyuan Chen, Yuxi Li, Mehrtash Harandi, Weiyao Lin

Neural Information Processing Systems (NeurIPS), 2025

Paper| Github

• Employ Diffusion Transformers as a feature extrator for visual correspondence!

🥑 Noise Learning

ACMMM 2025

DAC: 2D-3D Retrieval with Noisy Labels via Divide-and-Conquer Alignment and Correction

Chaofan Gan, Yuanpeng Tu, Yuxi Li, Weiyao Lin

ACM International Conference on Multimedia (ACMMM), 2024

Paper |Arxiv | Github

• Divide-and-Conquer stategy for 2D-3D multimodal noise learning!

📺 Video Understanding

NeurIPS 2024, Spotlight

MECD: Unlocking Multi-Event Causal Discovery in Video Reasoning

Tieyuan Chen*, Huabin Liu*, Tianyao He, Yihang Chen, Chaofan Gan, et. al

Neural Information Processing Systems (NeurIPS), Spotlight, 2024

Arxiv | Github

🎓 Educations

• 2022.09 - present, PhD Candidate (Joint), Information and Communication Engineering. Shanghai Jiao Tong University.
• 2022.09 - present, PhD Candidate (Joint), Electrical and Computer Systems Engineering. Monash University.
• 2018.09 - 2022.06, Undergraduate, Software Engineering. Huazhong University of Science and Technology.

📖 Publications

• C. Gan, Y. Tu, X. Chen, T. Chen, Y. Li, M. Harandi, W. Lin, “Unleashing Diffusion Transformers for Visual Correspondence by Modulating Massive Activations”, NeurIPS 2025.
• T. Chen, H. Liu, Y. Wang, Y. Chen, T. He, C. Gan, et al, “MECD+: Unlocking Event-Level Causal Graph Discovery for Video Reasoning”, ARXIV 2025.
• G. Zou, C. Gan, CH. Lim, S. Aramvith, W. Lin, “MCA: 2D-3D Retrieval with Noisy Labels Via Multi-Level Adaptive Correction and Alignment”, ICMEW 2025.
• T. Chen, H. Liu, T. He, Y. Chen, C. Gan, et al, “MECD: Unlocking Multi-Event Causal Discovery in Video Reasoning”, NIPS 2024, Spotlight.
• C. Gan, Y. Tu, Y. Li, W. Lin, “DAC: 2D-3D Retrieval with Noisy Labels via Divide-and-Conquer Alignment and Correction”, ACMMM 2024.