Shihong Zhang
| 📧 sh.zhang@tum.de | 📞 +49 15204777456 |  |
🌐 Personal Website |
Education
Technical University of Munich – School of CIT
Munich, Germany
04.2022 - Present
Master of Science in Robotics, Cognition, Intelligence
- Selected Coursework:
- Introduction to Deep Learning
- Computer Vision II: 3D Reconstruction
- Machine Learning
- Computer Vision III: Detection, Segmentation, and Tracking
Jilin University
Changchun, China
09.2016 - 06.2020
Bachelor of Science in Mechanical Engineering
- Thesis: Design and Dynamics Analysis of Auxiliary Belt Transmission System of HongQi (1.0)
Research Experience
Fast Diffusion Models
Seminar, TUM Info16 - Computer Aided Medical Procedures
04.2024 - Present
- Conducted an in-depth review of recent papers on fast diffusion models, summarizing key advancements.
- Key Topics Covered:
- UniPC: A plugin framework to enhance other samplers.
- DSNO: Achieves one-step sampling using Fourier convolution for temporal information capture.
- GET: Utilizes a DEQ model for distillation to enable one-step sampling.
- Summarized findings in a detailed blog on BayernCollab DLMA: Fast Diffusion Models.
Machine Learning in Crowd Modeling and Simulation
Practical Training, TUM Info5 - Scientific Computing
04.2024 - Present
- Trained a diffusion model using PyTorch to generate pedestrian trajectories, based on data collected from the MI building at TUM. The generated trajectories perfectly conform to the building’s layout.
- Engineered a Python-based crowd simulation tool utilizing cellular automaton to accurately track crowd parameters such as true velocity and flow rate. Developed a GUI for scenario visualization and designed highly modular code to support diverse environments.
- Orchestrated advanced crowd simulations with Vadere and integrated the SIR model to precisely model infectious disease spread, demonstrating the tool’s capability in realistic and complex scenarios.
End to End Autonomous Driving
Practical Training, TUM Info6 - Robotics, AI and Real-time Systems
10.2023 - 03.2024
- Trained two neural networks using PyTorch for end-to-end autonomous driving, leveraging Unity for data collection and simulation.
- Utilized ResNet50 to predict steering angles and control parameters from real-time images, and implemented YOLOv8 to detect vehicles for enhanced navigation in complex environments.
- Validated that the integrated control module of these networks enabled smooth and reliable driving in the simulated environment.
Semantic Segmentation with Self-Supervised Learning
Course Project, TUM Info9 - Computer Vision Group
04.2023 - 09.2023
- Conducted an object segmentation project using binary class semantic segmentation, employing pretrained embeddings and self-supervision techniques.
- Enhanced segmentation accuracy with pixel-adaptive convolutional nets and DINO self-supervised Vision Transformers.
Design and Dynamics Analysis of Auxiliary Belt Transmission System
Bachelor Thesis, JLU SMAE
08.2019 - 06.2020
- Developed a general-purpose static layout design software for belt drive system using VBA programming.
- Rated as excellent graduation thesis.
Automatic Mobile Phone Film Sticking Machine
Innovative Entrepreneurial Program, JLU SMAE
09.2018 - 06.2019
- Produced an automatic mobile phone film sticking machine.
- Obtained a patent, and published a paper in Science and Technology Vision.
- Awarded the second prize in Jilin Province.
Professional Experience
Physics and Mathematics Tutor, ZhangMen (China)
09.2020 - 10.2021
- Conducted personalized tutoring sessions, enhancing high school students’ understanding of physics and mathematics.
Skills
- Programming Languages and Operation Systems: Python, VBA, C++, C, MacOS, Linux, Windows
- Technical Tools: Git, Docker, Jekyll, LaTeX, PyTorch, Jupyter Notebook, Pandas, NumPy
- Languages: Chinese (Native Speaker), English (C1), German (B2)