Yihan Wang

A master's student at EPFL, working on 3D computer vision.

prof_yihan.jpg

I am a Master’s student at École polytechnique fédérale de Lausanne (EPFL), with a focus on computer vision and machine learning. I’m currently working as a research student focusing on parametrical diestortion free Structure-from-Motion at ETHz under the supervision of Prof. Dr. Marc Pollefeys in CVG. Previously, I spent some time in VITA lab conducting research on pose-estimation based lane marker detection, supervised by Prof. Dr. Alexandre Alahi. Prior to joining EPFL, my deep-seated interest in human-centric technology led me to conduct research on the spread mechanisms of respiratory diseases at Zhejiang University, CN, where I completed my bachelor’s degree, working under the supervision of Prof. Jianjian Wei.

My research interest broadly lie in the field of 3D computer vision, spanning over geometry related 3D mapping, localization, as well as deep learning based vision applications including but not limited to autonomous vehicles.

I always enjoy peaceful reading time, with favorite writers being Albert Camus and Franz Kafka, starting to discover the beauty of Kazuo Ishiguro. I’m also interested in documentary literature and featured articles from journalism. In my spare time, I love to explore local culture via traveling, and do sports like swimming, occasionally :smile:



Research Projects


sym

Structure-from-Motion with a Non-Parametric Camera Modell
Authors: Yihan Wang*, Linfei Pan*, Marc Pollefeys and Viktor Larsson
CVPR2025 Highlight

Webpage      Code      Video

A parametrical distortion free Structure-from-Motion (SfM) pipeline, featuring a partial calibration scheme for bootstraping a more complete calibration map. This novel method incorporates a much wider range of cameras into computer vision tasks. (i.e., fisheye and catadioptric).
sym

Simple Yet Efficient Monocular Lane Detection with Composite Fields Supervised by Reyhaneh Hosseininejad and Alexandre Alahi
Paper under submission

A novel scheme for keypoint-based lane detection using pose estimation, using composite fields to jointly estimate the keypoint locations and associations. This method achieves on par performance with state-of-the-art frameworks on public benchmarks while being relatively simple.

Course Projects


sym

A Chatbot for Commonsense Persona-grounded Dialogue Generation
Team members: Yihan Wang, Zewei Zhang,
Supervised by Silin Gao, Martin Jaggi and Nicolas Flammarion

A chatbot model generating dialogue responses that possess both persona consistency and contextual coherence.

Data Analysis and Visualization



sym

Analyzing User Preferences and Naming Impact for Guiding Targeted Recommendation
Team members: Siyuan Cheng, Zhixun Chen, Xinyi Ding, Yihan Wang,
Supervised by West Robert

A data story was created to analyze user preferences and naming impact using review datasets consists of both categorical ratings and textual reviews.
sym

Data-driven Analysis of City Scale Human-building Interaction
Supervised by Kanaha Shoji and Andrew Sonta

An analysis about human's interactions with urban forms based on pedestrian count data, with an aim to understand what kind of urban design is considered benefiting internal individual residences and cohesion of society.

Previous Publications

Prior to diving into computer vision, with the deep concern of the human wellfare, I have been working on the spread mechanisms of respiratory diseases. It led to a publication during COVID-19 pandemic.
sym

Tracing the origin of large respiratory droplets by their deposition characteristics inside the respiratory tract during speech
Yihan Wang, Jianjian Wei*, Caroline X. Gao, Tao Jin, Li Liu

Investigation on the importance of airborne transmission in the spread of respiratory infectious diseases using computational fluid dynamics based on a realistic human airway model.