Cong Yang

I am an Associate Professor at the School of Future Science and Engineering at Soochow University (SUDA), where I am heading the Ecology and Innovation Center of Intelligent Driving (BeeLab). My work is in the intersection of computer vision, machine learning and autonomous driving, with the goal of making autonomous driving system more robust, reliable and safe in various environments.

I was a Postdoc researcher at the MAGRIT team in INRIA (France). Later, I worked scientifically on computer vision and machine learning in Clobotics and Horizon Robotics. At Clobotics, I have worked on Smart Retail and Smart Wind with Dr. Yan Ke. At Horizon Robotics, I led the computer vision team and successfully delivered the intelligent cockpit (Changan UNI-T) based on Journey 2 SoC, making UNI-T the world's first massive-produced car using Chinese AI chips. I did my Ph.D. degree in computer vision and pattern recognition from the University of Siegen (Germany) in 2016, supervised by Prof. Dr. Marcin Grzegorzek.

Email  /  CV  /  Bio  /  Google Scholar  /  Chinese Page  /  Github

My research interests include skeleton extraction in images and shapes, image stitching, absolute and relative camera pose estimation, moire pattern detection and removal, fine-grained object classification, and vision-related perceptions in autonomous driving. My current work focuses on making algorithms on resource-limited SoC for environmental perception in autonomous driving scenarios.

• ✨ April 2024: I work as an editor in the special issue Recent Advances in Large Language Models in Electronics (IF=3.9). We are looking for your submission!
• ✨ April 2024: 1 paper accepted to International Conference on Multimedia Retrieval (ICMR 2024), source codes of YawnNet will be released soon!
• March 2024: I work as an editor in the research topic Efficient Algorithms for Bird's Eye View-based Perception in Frontiers in Signal Processing. We are looking for your submission!
• January 2024: 1 paper accepted to IEEE International Conference on Robotics and Automation (ICRA 2024), source codes of CAMA are available on GitHub.
• September 2023: 1 paper accepted to International Journal of Computer Vision (top journal, IF=19.5), source codes and datasets of SkeView are available on GitHub.
• August 2023: 1 paper accepted to CIKM 2023 (CORE A), congratulations to Zeyd Boukhers!
• July 2023: Source codes of our proposed RoMe (Large-scale Road Surface Reconstruction) is released in Github.

For an up-to-date list of my publications, please see my Google Scholar profile.

Lectures

• 2024-Winter: Machine Learning, Soochow University
• 2023-Winter: Computer Vision, Soochow University
• 2023-Summer: Computer Vision Practice, Soochow University
• 2023-Summer: Data Structure, Soochow University
• 2023-Summer: Data Structure Practice, Soochow University

Theses

• 2023:
  Shiyuan Chen: Towards Robust Vehicle and Pedestrian Detection via Monocular Camera, Tianjin Polytechnic University
  
• 2018:
  Oliver Tiebe: Automatic Skeleton Pruning for Graph-based Object Retrieval, University of Siegen
  

CAMA: Consistent and Accurate Map Annotation for Intelligent Driving.

In use at Horizon Robotics for 4D annotation

A simple yet efficient method, RoMe, for largescale Road surface reconstruction via Mesh representations.

In use at Horizon Robotics for 4D annotation

MoireDet algorothm for real-time Moiré Pattern Detection.

MoireScape dataset for training and evaluating moiré pattern detection and removal.

In use at Clobotics Smart Retail

FatigueView is a new large-scale dataset for vision-based drowsiness detection, which is constructed for the research community towards closing the data gap behind the industry.

In use at Changan UNI-T

Blade30 contains 1,302 real drone-captured images covering 30 full blades captured under various conditions (both on- and off-shore), accompanied by a rich set of annotations such as defects and contaminations, etc.

In use at Clobotics Smart Wind

BlumNet is a simple yet efficient framework for extracting object skeletons in natural images and binary shapes. BlumNet has significantly higher accuracy than the state-of-the-art AdaLSN (0.826 vs. 0.786) on the SK1491 dataset, a marked improvement in robustness on mixed object deformations, and also a state-of-the-art performance on binary shape datasets (e.g. 0.893 on the MPEG7 dataset).

SiTPose is a siamese convolutional transformer model to regress relative camera pose directly.

It uses odometry as input and estimates accurate ground plane normal vectors in real time.

In use at Horizon Driving Solutions

Pixels-IoU (PIoU) Loss is formulated to exploit both the angle and IoU for accurate oriented bounding box (OBB) regression.

In use at Clobotics Smart Retail

Retail50K is a collection of 47,000 images from different supermarkets. Annotations on those images are the layer edges of shelves, fridges and displays, for training and evaluating oriented bounding box (OBB) detectors.

WatchPose is a simple yet efficient camera pose data collection method to improve the generalization and robustness of camera pose regression models.

Source codes of 17 contour segment (CS) descriptors and 4 CS datasets.

A 3D object matching framework based on stripes generated from laser scanning lines.

Editor:

• Frontiers in Signal Processing: 2022

Program Committee:

• International Workshop on Sensor-Based Activity Recognition and Artificial Intelligence: 2023

Workshop Chair:

• International Conference on Cybernetics: 2017

Reviewer:

• ACM Multimedia
• IEEE International Conference on Image Processing
• International Journal of Computer Vision
• IEEE Transactions on Visualization and Computer Graphics
• IEEE Transactions on Industrial Informatics
• IEEE Transactions on Intelligent Transportation Systems

This website is based on the source code of Jon Barron's website.