Yingwei Song

Yingwei Song Hi there, I am a current MSCS student at Brown University. I completed my bachelor's degree in Computer Science and Data Science from UW-Madison. I am interested in computer vision, machine learning and robots controlling, especially image understanding.

---

Research and projects:

Image Synthesis with Transformer: Brain MR Image to PET Image
Proposed a novel cycleGAN framework to generate brain PET image from MR image. Paper is still ongoing.
[paper][github]

Autonomous & Resilient Controls Lab, UW-Madison
Implement MPC controller algorithms on Mushr Car.
[[demo video](https://youtu.be/JYFgtiQO8VE"demo video”)][github](Github access may be denied)

Computational Optics Group, UW-Madison
Implement and optimize the deblurring algorithm for images detected by SPAD(Single Photon Avalanche Diodes).
[final report]

Simulation of the Connected and Automated Driving Systems UW-Madison
Design maps using RoadRunner, design detailed traffic scenarios; apply the planning algorithms to do the microscopic and macroscopic route planning and find the optimal movement trajectory for the vehicles(PID controller).
[github]

Rural Housing in Henan, China: Drone Photography Area Estimation based on Deep Learning
Image segmentation and area estimation based on UNET.

Projects Title

Computational Optics Group - Adaptive Motion Deblurring

Location

University of Wisconsin-Madison, USA

Time

2022.8 - present

Advisor

Andreas Velten, PH. D, Assistant Professor at the Department of Biostatistics and medical Informatics, Electrical and Computer Engineering

Trevor Seets, NSF Graduate Research Fellow at Computational Optics Lab

Description

Compressing the camera image can always a problem in real world. Due to SPAD’s unique ability to precisely time-tag individual photons, it has great potential for high-resolution long-range LiDAR systems. We pick the series images captured by SPAD and select change points(by solving optimal problem). Then uses the change points to simulate the original object’s real movement to deblur. Also, we find another way to simulate real images only based on the changes point images, which can speed up the calculation.

Implementation

Mainly focused on adjusting the deblurring methods used and visualizing results; detecting change points and the deblurred image based on change point videos; also using novel interpolation to optimize the deblurring method.

LINK:https://github.com/wrencanfly/Adaptive_Motion_Deblurring (permission may be required)

Projects Title

Autonomous & Resilient Controls Lab - Mushr Car implementing MPC algorithm

Location

University of Wisconsin-Madison, USA

Time

2022.9 - present

Advisor

Xiangru Xu, PH. D, Assistant Professor in the Department of Mechanical Engineering

Description

Discrete Model Predictive Control is implemented on Mushr Car, and Acado Toolkit is used to solve the optimization problem. The Mushr Car’s position and orientation are obtained from the ROS and passed as input to the generated Acado Solver, and the velocity and steering angle output from the solver are passed to the ROS to control the car.

Implementation

We used the OptiTrack to get the Mushr Car’s real position and passed the information back to the offline Acado Solver on the Mushr Car. Several preset paths are successfully implemented on Matlab, and the Figure-8 path is tested on the real Mushr Car, which is demonstrated in the above video.

LINK: https://github.com/wisc-arclab/arclab_vehicles/tree/mushr (permission may be required)

Projects Title

Image Synthesis with Transformer: Brain MR Image to PET Image

Location

University of Wisconsin-Madison, USA

Time

2022.9 - present

Advisor

Vikas Singh, PH. D, Professor in the Department of Biostatistics

Yong Jae Lee, PH. D, Associate Professor in the Department of Computer Sciences

Description

To reduce the financial burden on the patient and to speed up the diagnostic process, we propose a novel unsupervised GAN framework to achieve the goal of using magnetic resonance imaging (MRI), an inexpensive and easy-to-use technique, to generate certain types of images of positron emission to mography (PET), which is an expensive and complex time-consuming technique.

Implementation

Our key idea is to use CycleGAN to map the input image to a specified one in the output domain and use a combination of MobileNetV2 and self-attention in the process to achieve global and local feature capture of the input photo. We present experiments on the Alzheimers Disease Neuroimaging Initiative (ADNI) dataset with 726 training and 80 testing subjects and obtain acceptable per- formance in PET image synthesis. We also use various metrics to evaluate the generated images.

GitHub LINK: https://github.com/GeofrreyLi/U-TransCyGan

Paper draft(on going):

**Related study:

Swapping Auto-Encoder presentation: Slides

Projects Title

Simulation of the Connected and Automated Driving Systems (Collaborative Automated Driving System)

Location

University of Wisconsin-Madison, USA

Time

2021.10 - present

Advisor

Bin Ran, PH. D, Vilas Distinguished Achievement Professor

Yang Cheng, PH. D, Transportation Research Associate

Description

The Collaborative Automated Driving System (CADS) provides full vehicle operations and control for Connected and Automated Vehicle and Highway (CAVH) systems, which include
an intelligent vehicle subsystem and an intelligent roadway subsystem, by sending individual vehicles with detailed and time-sensitive control instructions for vehicle operations. The
Connected Reference Marking System (CRMS) and related methods provide vehicle location identification functions for connected automated vehicles (CAV) at any vehicle intelligence
level.

Implementation

• The Map and Scenario Design Group: design maps using RoadRunner, design detailed traffic scenarios; test the maps and scenarios in CARLA; collaborate with other groups to do the simulation
• The Planning and Control Group: Apply the planning algorithms to do the microscopic and macroscopic route planning and find the optimal movement trajectory for the vehicles.

Current process:

Github LINK:

https://github.com/wrencanfly/carla_wisc_PlanGroup

CARLA Build Document:

https://www.yuque.com/beichenandjojo/csgo/fz4w48?singleDoc# 《CARLA BUILD WINDOWS》

Projects Title

Area Division of Rural Housing in Henan Province: Drone Photography Area Estimation based on Deep Learning

Location

Henan University

Time

2019.10 - 2020.1

Advisor

Haiying Wang, Associate Professor

Aim

Use Pytorch to realize the UNET algorithm to build a ResNet50 neural network to implement the image segmentation of house roofs.

Implementation

• Wrote code to divide unprocessed drone image into 81 equal sections and applied “labelme” to mark 200
  image training datasets
• Conducted image enhancement using OpenCV to facilitate machine learning
• Configured all needed environments and called the training backbone built by Russian software developer
  Pavel Yakubovskiy to realize the UNET algorithm
• Built the main structure using Python
• Tested the outcomes on n unlabeled images

Outcomes

​ This is the first day I built my own website. Actually, it is not my first website. I guess 10 years before maybe - when I was still in middle school. At that time, I use BAE(a free platform supported by Baidu) and the domain with chogellge.ml as my first website. But at that time, I totally use WordPress and I followed the instructions online step by step. That’s super easy to build a website, and I remembered that very few problems I met. And everything would be super easy and fancy when using WordPress. That also might be because I really had. enthusiasm on this. And I can’t believe I really have my website at that time, though I have nothing to record and post when I was a middle-school student.

​ Now it almost passed ten years, and things changed a lot. When I was a kid, I don’t know what is frontend, backend, and any algorithms, I have no idea about ML and data science. At that time, I believed that making my own website was really cool and it was a thing I can show off to girls(…that is what exactly I’ve done before).

​ Time flies, now, I never think deploying a website is really a thing I can show off, the thing is, I actually do the same thing as ten years ago - I followed the instructions step and step…(the only thing I do better than before is, I am more familiar with is Linux commands, git commands, and English!!)

​ The reason I build up my own website is to show some of my interests, researches, and things I want to write. I will try to write everything here in English - which is apparently not my native language, and I will try to minimize the grammar mistakes and make things logically - but I cannot make sure! Cuz this is not going to be viewed by my ESL(English for secondary language) course teachers!