Education

Master in Biomedical Computing @TU München, Informatik

Oct/2009 - Apr/2012 GPA 1.8

  • 3D Computer Vision, Optimization, Machine Learning.

  • System and Statistical Model Generation for X-ray Computed Tomography (Note 1.3)

Bachelor in Biomedical Engineering @Shanghai Jiao Tong University

Sep/2005 - Jul/2009

  • Wireless ECG Signal Monitoring and Processing System (Bachelor Thesis)

  • Accepted without entrance examination for the 1st Prize of Chinese Physics Olympiad

Work Experience

Machine Learning Engineer @ESR Labs/Accenture

Oct/2019 - present

  • Hardware-aware deep neural network design and compression.

  • Neural network based lane detection. Lane estimation and tracking with prior knowledge fusion.

  • Semi-automatic and automatic ground truth annotation for different tasks.

  • Software architecture design by considering company's market position, business use cases, technical strength and suppliers' limitations.

  • Continuous semantic grid map construction and fusion.

Group Lead @Hyundai Mobis

Oct/2018 - Oct/2019

  • Lead a development group for motion estimation related features of the autonomous driving software stack, including vision odometry, sensor fusion, online camera calibration etc.

  • Involved in the production project as a lead, software architect, algorithm designer, and feature developer.

  • Development of a deep learning framework for geometrical tasks(camera pose, mono depth and optical flow).

Computer Vision Engineer @Hyundai Mobis

Oct/2017 - Oct/2018

  • Constructed a multi-stage visual odometry framework supports both optimization and filtering.

  • Development of a deep learning framework for geometrical tasks(camera pose, mono depth and optical flow).

Software Developer @e.Sigma Systems GmbH

May/2012 - Sep/2017

  • Designed and developed a recognition and tracking based virtual shooting simulator

  • Designed and developed a 3D object calibration and real-time pose estimation framework

  • Embedded Linux-based tracking system, responsible for the software part, including algorithm design and coding

  • Network-based distributed simulation engine, worked in a big team, responsible for some modules, including core modules

  • Ultra-low latency real-time distributed system, low-level Linux kernel development, optimzation of a big system

Research Assistant @Klinikum Rechts der Isar

Jul/2010 - Apr/2012

  • 3D CT reconstruction framework and algorithm development

Research Assistant @Lehrstuhl CAMP/AR, Fakultät Informatik, TU München

Jun/2010 - Feb/2011

  • Human Airway Segmentation from 3D CT Image

Research Assistant @NarvisLab, Lehrstuhl CAMP/AR, TU München

Feb/2010 - Aug/2010

  • Phase Aberration Correction for Ultrasound Image Reconstruction

Projects

Environment Perception of an Autonomous Driving System@ESR Labs/Accenture

C++/Tracking/Grid Map Fusion/Lane Estimation/PyTorch/TensorRT

  • Responsible of developing lane detection neural network from data annotation to target hardware deployment
  • Continuous statistical fusion of semantic grid map based on camera semantic segmentation
  • Designed and developed a continuous lane estimation and tracking system which supports prior knowledge fusion

Object Detection Prototype on an Embedded Platform@ESR Labs/Accenture

C++/Hardware Aware Network Design/Neural Network Compression

  • Designed and trained an object detection network based on the consideration of the hardware's strength and limitations
  • Developed a neural network compression application which reduces the inference time by 50%
  • Designed and developed a complete object detection solution. Implemented on an embedded platform from camera driver to video output, the system achieves high accuracy and runs at 60 fps

Neural Network Pipeline @ESR Labs/Accenture

C++/TensorFlow/PyTorch/TensorRT/Deep Learning

  • Designed and developed a semi-automatic lane annotation application
  • Implemented a fully automatic object detection and instance segmentation annotation system
  • Designed and implemented a C++ neural network library on an embedded platform, the library supports 3 different neural network inference frameworks

Architecture Design of a Distributed Event-driven Data Collection System@ESR Labs/Accenture

Market Analysis/Requirement Engineering/System Design

  • Requirements engineering and business analysis based on the client's market position
  • Clarification of the system limitations by directly communicating with the client's hardware and software suppliers
  • Completed the software architecture design, which covers the whole system from cloud to subsystems' flowcharts on edge device

Autonomous Vehicle Motion Estimation @Hyundai Mobis

C/BLAS/CMake/OpenCV/TensorFlow/SLAM/Deep Learning

  • Implemented a visual odometry framework in pure C/Blas, 5x faster than OpenCV with higher accurary
  • Designed and implemented an adaptive Kalman-Filter based sensor fusion framework for vehicle motion estimation
  • Designed a novel CNN based feature points selection schema, under development
  • Research, investigate and test the state-of-the-art CNN based camera pose estimation and depth map reconstruction algorithms

Image Series based 3D Reconstruction @e.Sigma Systems GmbH

OpenCV/Qt/OpenGL/Camera Calibration/3D Reconstruction

  • Implemented a camera extrinsic calibration algorithm based on image series
  • Designed and implemented a semi-automatic interactive 3D reconstruction pipeline
  • Constructed a 3D modelling environment which supports creating, editing, saving, reading and rendering 3D scene
  • Released the software as a product with a USB-Dongle based licence

3D Object Pose Estimation @e.Sigma Systems GmbH

Matlab/Optimization/Pose Estimation

  • Developed an API to model and calibrate a real word object for a virtual-reality environment
  • Developed a template-matching algorithm to estimate the pose of an object in real time

Shooting Simulator @e.Sigma Systems GmbH

Tracking&Recognition/Machine Learning/Embedded Linux
Camera Programming/Camera Calibration

  • Developed a calibration algorithm for multi-camera system
  • Implemented a multi-target tracking algorithm by fusing camera and sensor data
  • Designed and constructed a tracking system exchanging data via Network

3D CT Image Reconstruction @Klinikum Rechts der Isar

Machine Learning/Optimization/Image Reconstruction

  • Proposed a mathematical model of a commercial medical CT
  • Built a Poisson distribution based probabilistic reconstruction framework
  • Developed an optimization algorithm to raise the reconstruction convergence speed
  • Achieved a better reconstruction quality than the most advanced commercial CT from the same data set

Automatic Image Stitching @e.Sigma

Computer Vision/Image Processing

  • Implemented an automatic image pairing and stitching algorithm
  • Developed an image blending algorithm based on brightness and contrast adjustment

Real-Time Distributed Data Acquisition System @e.Sigma Systems GmbH

Hardware Driver/Linux Kernel/Real-Time System/
Multithreaded Programming

  • Optimized the driver of board hardware on Real-Time Linux
  • Improved a whole distributed real-time system by optimizing its scheduling strategies
  • Developed the DAS’s Linux kernel modules

Distributed Simulation Engine @e.Sigma Systems GmbH

Game Engine/Network Programming

  • Optimized the network-based data exchange core of a distributed simulation engine
  • Developed some plugins and modules for a distributed simulation engine
  • Developed the driver for a HID device to drive plane in a simulation engine

RFID Sensor based Localization @e.Sigma Systems GmbH

Sensor Data Processing

  • Developed an API to monitor RFID sensors and locate them