Haodong Li

I am a research intern at University of Pennsylvania, supervised by Prof. Lingjie Liu, and a M.Phil. student at Hong Kong University of Science and Technology (Guangzhou Campus), supervised by Prof. Ying-Cong Chen and Prof. Xin Tong. Previously, I received my B.Eng. degree in Automation from Zhejiang University in 2023.

My research interests include 3D vision and Generative Models. My personality type is ENTJ. Feel free to drop me an email if you have anything to discuss!

I am actively seeking for Ph.D. positions for Fall 2025!

Email  /  CV  /  Scholar  /  Twitter  /  Github

Research

arXiv 2024
arXiv / Project Page / Github (Soon)

Video depth estimation is not merely an extension of image depth estimation. The consistency for dynamic and static regions in videos are fundamentally different. To tackle these challenges, StereoDiff synergizes stereo matching with video depth diffusion models, achieving superior video depth estimation performance.

arXiv 2024
arXiv / Project Page / Github / Demo (Depth) / Demo (Normal)

Lotus is a diffusion-based visual foundation model with a simple yet effective adaptation protocol, aiming to fully leverage the pre-trained diffusion's powerful visual priors for dense prediction. With minimal training data, Lotus achieves SoTA performance in two key geometry perception tasks, i.e., zero-shot monocular depth and normal estimation.

arXiv 2024
arXiv / Project Page / Github / Demo (Soon)

Characterized by its emphasis on the interpretation of subject-essential attributes, the proposed DisEnvisioner effectively identifies and enhances the subject-essential feature while filtering out other irrelevant information, enabling exceptional image customization without cumbersome tuning or relying on multiple reference images.

arXiv 2024
arXiv / Project Page / Github

DOME is a diffusion-based world model that predicts future occupancy frames based on past observations. DOME exhibits two key features: (1) high-fidelity and long-duration generation enabled by its spatial-temporal diffusion transformer; and (2) fine-grained controllability, thanks to the trajectory encoder.

SIGGRAPH Asia 2024

DIScene is capable of generating complex 3D scene with decoupled objects and clear interactions. Leveraging a learnable Scene Graph and Hybrid Mesh-Gaussian representation, we get 3D scenes with superior quality. DIScene can also flexibly edit the 3D scene by changing interactive objects or their attributes, benefiting diverse applications.

CVPR 2024 Highlight
arXiv / Paper / Github / Demo / Video

We present LucidDreamer, a text-to-3D generation framework, to distill high-fidelity textures and shapes from pretrained 2D diffusion models with a novel Interval Score Matching objective and an advanced 3D distillation pipeline. Together, we achieve superior 3D generation results with photorealistic quality in a short training time.

ECCV 2024
arXiv / Paper / Project Page

We introduce Bi-TTA, a method that leverages spatial and temporal consistency for appropriate self-supervision, coupled with novel prospective and retrospective adaptation strategies, enabling superior adaptation ability of pre-trained rPPG models to the target domain using only unannotated, instance-level target data.

Academic Service

Reviewer: ICLR 2025

Education