Wenbo Ji   嵇jī / yí文博

I am a second year master's student in Electrical Engineering and Information Technology at the Technical University of Munich.

I am doing research internship with Jiapeng Tang and Prof. Matthias Nießner at TUM Visual Computing Group.

I was working as a working student at ImFusion with Mert Asim Karaoglu, Dr. Benjamin Busam, Prof. Nassir Navab.

I was a master's student of TUM-CVG, TUM-DI-LAB working with Prof. Daniel Cremers, Dr. Yan Xia, Dr. Chuanxia Zheng and Weirong Chen.

I did a research internship at Zhejiang University under the supervision of Prof. Yiyi Liao.

I obtained my Master's degree in Computer Science from the Tongji University.

I did my B.Sc. in Information and Computing Science at Nanjing Tech University.

My research interests focus on moving beyond isolated visual subproblems toward integrated models of virtual humans and 3D scenes that can perceive, generate, and reason about interaction in dynamic environments.

中文 / English / Deutsch

I'm looking for PHD positions starting in 2027.

• [10/2025] Our paper CSG-Fusion received the 🏆 Best Paper Award at the ICCV 2025 Workshop E2E3D!


• [08/2025] Our paper CSG-Fusion is accepted by ICCV 2025 Workshop E2E3D.


• [07/2025] I graduated from Tongji University with a Master's degree in Computer Science! 🎓


• [06/2025] Our paper LiteTracker is accepted by MICCAI 2025.


• [01/2025] Our paper Re0 is accepted by ICRA 2025.

---- show more ----
[08/2024] I will join Imfusion as a working student on the topic of Dense Video Point Tracking.


[01/2024] I will join TUM-CVG, TUM-DI-LAB under the co-supervision of Dr. Yan Xia and Dr. Chuanxia Zheng.


[05/2023] I will join Technische University München in the winter semester 23/24 as a Master student in Electrical Engineering, see you in München🍻!

Long-term Tracking 3D/4D Reconstruction Scene Decomposition Video Generation Avatar Animation Human Scene Interaction

My previous work on 3D reconstruction, scene decomposition, and tracking forms a coherent progression toward a broader goal: developing unified visual models that move beyond explicit intermediate representations and support generation and interaction in dynamic environments.

* equal contribution, † corresponding author

ICCV Workshop E2E3D, 2025 Best Paper Award

A feed-forward framework that mindfully integrates pixel-aligned pointmap to reduce redundant primitives and produce compact and consistent 3D structures.

MICCAI, 2025

A low-latency method for tissue online tracking in endoscopic video streams, introduces a set of training-free runtime optimizations on a state-of-the-art long-term point tracking method.

ICRA, 2025

Given 3D point clouds and multi-view RGB-D images with poses, Re0 leverages the 3D geometric information, projection relationships and CLIP semantic features for 3D zero-shot instance segmentation.

Given stereo endoscopic videos of deformable tissues, this thesis proposes a 4D deformable half-Gaussian splatting framework with depth-prior initialization, temporal–spatial encoding, and lightweight Gaussian pruning for high-fidelity, real-time surgical scene reconstruction.

Some of my 3D artwork and renders created with Blender.