Weize Li (李维泽)

I am currently spending my gap year as a research intern student at Institute for AI Industry Research (AIR), Tsinghua University. I am advised by Prof. Hao Zhao from DISCOVER Lab. Previously, I was a visiting student at Institute of Automation, Chinese Academy of Sciences in my senior year. And I obtained the Bachelor degree in Mechatronics Engineering from Beijing University of Civil Engineering and Architecture, advised by Prof. Miao Yu.

I am a active member of AnySyn3D, a non-profit research interest group comprising individuals with a strong interest in exploring research problems and cutting-edge technologies in any topics of 3D.

I am actively looking for the PhD opportunity in Fall 2025.

Email  /  CV  /  Google Scholar  /  GitHub  /  Twitter

My long-term research goal is to build embodied agents that can effectively understand 3D worlds based on vision and be able to combine visual concepts and external knowledge for interpretable reasoning. Further, being able to apply the learned knowledge from reconstruction to generation and editing the digital world, towards the ultimate goal from understanding to changing the world. So I am up for anything related research and currently focusing on:

We introduce the new task of outdoor 3D dense captioning with TOD3Cap dataset; We propose TOD3Cap network, leveraging the BEV representation to encode sparse outdoor scenes, and combine Relation Q-Former with LLaMA-Adapter to dense captioning in the open-world.

We introduce pose-agnostic setting to 3D-aware object anomaly detection problem with MAD dataset, propose the NeRF-based anomaly detection framework: OmniposeAD.

We propose a hybrid model of improved RF + DNN to extract time-series features and classify anomalies in power system data.

This is an ongoing survey paper and we are conducting a systematic classification and in-depth analysis for 3D content editing based on radiance field representations (including NeRFs, 3DGS, etc.).

Investigated one‑image‑to‑3D methods like Zero1‑to‑3 to enhance the quality of multi‑view car instances in simulation scenes, with a focus on fine‑tuning models using the ”Car” labeled 3D assets from the Objaverse dataset.

This project proposes a unified framework for visual anomaly detection based on heterogeneous knowledge distillation. The One-model-all class framework merges CNN and ViT with class-specific learnable query to enable mutually facilitated learning of anomalous features across multi-class samples.

I served / was delegated as Reviewer for NeurIPS’23, CVPR’24.

Outside of research, I enjoy playing football⚽, fitness💪 and photography📷. I am a member of the Certified Referee🗣️ Crew of the Chinese Football Association. I am also a big fan of Taylor Swift🦋.

Website template from Jon Barron.