Zhangyu Jin

I am a first-year PhD student (2024 Fall-) in Computer Science at the University of Southern California, supervised by Prof. Mohammad Soleymani and Dr. Wei-Wen(Andrew) Feng.

I received my Master’s degree from University of Southern California. During my Master’s, I took Research Internships at Motorola Solutions and at TuSimple. Previously, I was a Research Engineer at YITU Technology for one year. Prior to that, I got my Bachelor degree from Shanghai Jiao Tong University.

My research interests mainly lie in Video Perception and Generative Models.

selected publications

Lossless Face Anonymization on Downstream Tasks

Zhangyu Jin, Siddharth Agrawal, and Yanyan Hu

Nov 2023

Abs HTML

Completed Face Anonymization by substituting each face with a similar one from the fakely generated face gallery, effectively hiding facial information from all users. Utilized as both the training and testing data to guarantee no decline in performance for all downstream tasks (ReID, Object Detection).
MemoryFlow: Efficient Video Object Detection

Zhangyu Jin, Haiping Wu, and Sha Long

Aug 2022

Abs HTML

Innovated MemoryFlow for efficient Video Object Detection, integrating neighboring optical flow content and long-term sequential content from memory. Reached 2nd place in ImageNet-VID ranking under ResNet-50 with fewer reference information and a faster inference time.
Real-time Panoptic Segmentation for Autonomous Driving

Zhangyu Jin, Yikai Hua, and Wenbin Lu

Jul 2021

Abs HTML

Incubated the Panoptic Segmentation for Autonomous Vehicles from scratch, providing prior information for all downstream tasks. Accomplished training-based Model Compression to reduce inference time, resulting in YITUTech’s successful completion of the first road test.