Muheng Li

Currently, I am a joint doctoral student at Paul Scherrer Institut (PSI) and ETH Zurich, supervised by Prof. Antony Lomax and Dr. Ye Zhang. Before, I graduated with a M.S degree from the Department of Automation at Tsinghua University, where I was advised by Prof. Jiwen Lu and Prof. Jianjiang Feng. I also had my bachelor's degree at the department of Engineering Physics, Tsinghua University.

Now, I am working at PSI’s Center for Proton Therapy, where I am deeply involved in the intersection of AI technology and radiation therapy. The core of my research is focused on addressing the complex challenges associated with 4D data analysis and modeling.

The dynamic nature of organs in motion during therapy sessions produces intricate 4D data. In my earlier academic journey, my research was concentrated on computer vision and deep learning, specifically generative 3D vision and multi-modal video learning. Building on my background, I am dedicated to advancing the methodologies for 4D data analysis, aiming to enhance the precision and effectiveness of proton therapy for cancer treatment.

Email  /  Google Scholar  /  Github

Publications

* indicates equal contribution

We propose a new generative 3D modeling framework called Diffusion-SDF for the challenging task of text-to-shape synthesis. We propose a SDF autoencoder together with the Voxelized Diffusion model to learn and generate representations for voxelized signed distance fields (SDFs) of 3D shapes. We also extend our approach to further text-to-shape tasks including text-conditioned shape completion and manipulation.

We propose a vision-language prompt-based framework, Bridge-Prompt (Br-Prompt), to model the semantics across multiple adjacent correlated actions, so that it simultaneously exploits both out-of-context and contextual information from a series of ordinal actions in instructional videos.

We propose an uncertainty-aware representation Learning (UARL) method for action segmentation. Specifically, we design the UARL to exploit the transitional expression between two action periods by uncertainty learning.

We propose a weakly-supervised approach called Order-Constrained Representation Learning (OCRL) together with a special contrastive loss function called StepNCE to predict future actions from instructional videos by observing incomplete steps of actions.

Website Template