Hi there! My name is Shi Chen (Simplified Chinese: 陈实, Japanese: 陳 実). I am a first-year PhD student at the Mobile Robotics Lab (MRL) of ETH Zurich, advised by Prof. Dr. Stefan Leutenegger. Previously, I completed my bachelor’s degree at Kyoto University, where I finished my thesis under the supervision of Prof. Ko Nishino and Prof. Shohei Nobuhara, and my master’s at ETH, where I worked mainly with Dr. Danda Pani Paudel and Prof. Martin R. Oswald.

My research interests lie in 3D vision, with a particular focus on reconstruction, understanding and reasoning of dynamic scenes. At MRL, I aim to push the boundaries of Spatial AI by endowing robots with physics-aware perception of dynamic environments.

Education

• 2025 - Present, Ph.D. Mechanical and Process Engineering, ETH Zurich, Switzerland.
• 2022 - 2025, M.Sc. Electrical Engineering and Information Technology, ETH Zurich, Switzerland.
• 2018 - 2022, B.E. Electrical and Electronic Engineering, Kyoto University, Japan.

Selected Projects

ProDyG: Progressive Dynamic Scene Reconstruction via Gaussian Splatting from Monocular Videos

Master’s thesis and NeurIPS 2025
Advisors: Dr. Erik Sandström, Dr. Sandro Lombardi, Siyuan Li, Prof. Martin R. Oswald
[arXiv][Project Page]

• Online dynamic scene reconstruction pipeline using 3D Gaussian Splatting from unposed monocular videos.
• Yields novel view renderings competitive to offline methods and achieves on-par tracking with state-of-the-art dynamic SLAM methods.

MonoDy-GS: Online Monocular Dynamic Gaussian Splatting

Semester project in Spring 2024
Advisors: Prof. Martin R. Oswald, Dr. Sandro Lombardi, Prof. Marc Pollefeys
[Slides][Report]

• An online system using 3D Gaussians to model dynamic scenes from monocular RGB-D inputs.
• Incorporated various real-world priors to compensate for the lack of multiview information.

EvenNICER-SLAM: Event-based Neural Implicit Encoding SLAM

Semester project in Fall 2022
Advisors: Dr. Danda Pani Paudel, Prof. Luc Van Gool
[Slides][Report][arXiv][Code]

• Integrated event input into the NICE-SLAM pipeline to for more robust camera tracking in challenging scenarios involving fast camera motion or low frame rates.
• Quantitative evaluations demonstrate that EvenNICER-SLAM significantly outperforms NICE-SLAM in scenarios with reduced RGB-D input frequency.

NeRaser: NeRF-based 3D Object Eraser

Course Project for Mixed Reality
Shi Chen, Liuxin Qing, Shao Zhou, Xichong Ling
Advisors: Dr. Sandro Lombardi, Prof. Marc Pollefeys
[Demo][Poster][Report][Code]

• Built upon nerfstudio an interactive framework for object removal from 3D NeRF-represented scenes.
• Applied a 3D-aware inpainting strategy to ensure multiview visual consistency.

Hierarchical Dense Neural Point Cloud-based SLAM

Course project for 3D Vision
Shi Chen, Guo Han, Liuxin Qing, Longteng Duan
Advisors: Dr. Erik Sandström, Prof. Martin R. Oswald
[Poster][Report][Code]

A comparison of the resulting neural point cloud reconstructed from ScanNet scene 0181.
(Left: Point-SLAM. Right: Ours)

• Built upon Point-SLAM a coarse-to-fine-optimization strategy leveraging multiple sets of point cloud with varying resolutions.
• Improved tracking robustness against real-world imaging effects such as motion blur and specularities.

Monocular Visual Odometry

Course Project for Vision Algorithms for Mobile Robotics
Shi Chen, Bowei Liu, Hanyu Wu, Kehan Wen
Supervisor: Prof. Davide Scaramuzza
[Demo(KITTI)][Demo(Malaga)][Demo(Parking)][Report]

• Implemented in MATLAB a monocular feature-matching-based visual odometry pipeline.

Road Scene Pedestrian Relocation for Data Augmentation

Bachelor’s thesis
Supervisors: Prof. Ko Nishino, Prof. Shohei Nobuhara
[Slides][Thesis]

• Devised data augmentation method that automatically cuts out large-scale pedestrians in foreground of road-scene videos and relocates them at farther positions with correct scale and occlusion.
• Proposed method improves Mask R-CNN in both detection and instance segmentation of far-away pedestrians.

Miscellaneous

I was born into a Chinese family in Yokohama, Japan, and spent my childhood there, so I am fluent in both Mandarin Chinese and Japanese. I can also speak some (洋泾浜) Shanghainese and (エセ) Kansai dialects.

I’m a big fan of Karaoke and J-POP music. My favorite artists include BUMP OF CHICKEN, Hikaru Utada, Tokyo Incidents, Kenshi Yonezu, King Gnu, Mr.Children, Sakanaction, Gen Hoshino, and TOMOO. It can take forever if you ask me for J-POP recommendations.

I enjoy watching football (soccer), basketball and tennis games. Not going to reveal here which teams I support, just in case you happen to root for their rival teams.