CV

Jianwen Cao

jianwen.cao@uzh.ch
+41 766395217
Zurich, , CH

Summary

M.Sc. Student in Informatics (AI) at University of Zurich (UZH). Interested in Computer Vision, Robotics, and Multimodal Models.

Education

  • Informatics (Major: Artificial Intelligence, Minor: Data Science)
    Present
    University of Zurich (UZH)
    GPA: 5.57/6
    Courses: Data Science, Machine Learning for Natural Language Processing, Vision Algorithms for Mobile Robotics, ETHZ 3D Vision, ETHZ Virtual Reality
  • Artificial Intelligence
    2024-06-01
    Nanjing University of Information Science and Technology (NUIST)
    GPA: 87.6/100

Work Experience

  • Student Researcher
    2025-03-01 - 2026-02-01
    Computer Vision and Geometry Group (ETHZ CVG)
    Reproduced and improved GS-LIVO to achieve better geometric accuracy and mapping efficiency, reaching SOTA performance.
  • Student Researcher
    2025-02-01 - 2026-01-01
    Robotics and Perception Group (UZH RPG)
    Conducted autoregressive pre-training on multimodal data, significantly improving the performance and robustness of downstream vision and control tasks.
  • Computer Vision Algorithm Intern
    2023-07-01 - 2023-10-01
    IntSig Information Co., Ltd.
    Developed object detection algorithms based on improved DETR series architectures.

Skills

Robotics

  • Robot Arm
  • MuJoCo
  • RL/IL
  • ROS2

Computer Vision

  • SLAM
  • 3DGS
  • Object Detection
  • DETR
  • GS-LIVO

Models

  • VLA
  • LLM
  • Generative Models
  • Diffusion
  • Multimodal

Publications

  • Generative Event Pretraining with Foundation Model Alignment
    2026
    CVPR
    Proposed a multimodal pre-training method combining images and event cameras, achieving SOTA performance.
  • Memory Over Maps: 3D Object Localization Without Reconstruction
    2026
    IROS
    Responsible for 3D object localization (query input, 3D position output) and efficiency analysis. Ultimately deployed on the Boston Dynamics Spot robot dog.
  • Fuse Tune: Hierarchical Decoder Towards Efficient Transfer Learning
    2023
    PRCV
    Proposed an efficient transfer learning method that eliminates the need for backbone network backpropagation.

Portfolio

  • Real-time Visual-Inertial-LiDAR Gaussian Splatting SLAM
    Portfolio
    Reproduced and improved GS-LIVO to achieve better geometric accuracy and mapping efficiency, reaching SOTA performance.
  • Event Camera-Enhanced Robot Motor Control Policies
    Portfolio
    Conducted autoregressive pre-training on multimodal data, significantly improving the performance and robustness of downstream vision and control tasks.