Zongwen Mu
Perception Engineer @Autox
Signup · Get unlimited contacts
WORK HISTORY
Perception Engineer @Autox
Developed time-sequential deep learning model for traffic light classification using ResNet as backbone, Transformer with auxiliary supervision as neck, achieved a 0.81 mAP on customized dataset. Designed clip sampler to generate sequence labels from single frames dataset, used multi-processing to accelerate the processing time from over 400s to 90.52s. Introduced class-balanced cross entropy loss to resolve the unbalanced data distribution issue, visualized test results for bad cases to help discovering annotation mistakes. Refactored the traffic light detection pipeline, optimized its basic logic, improved its process time by over 20ms, introduced new features to handle difficult edge cases. Collaborating with QA team, troubleshooting for traffic light pipeline, improving its performance and stability in onboard usage. Maintaining the deep learning models in traffic light pipeline, fine-tuning these models regularly, improved detection and classification accuracy by 10%.
EDUCATION
Carnegie Mellon University
Master of Science - MS
Zhejiang University
Bachelor's degree
ABOUT ZONGWEN MU
Perception engineer at AutoX.ai. Working on the traffic light detection pipeline with both deep-learning models and onboard code for self-driving cars. I received my degree of M.S in Mechanical engineering at Carnegie Mellon University, and I\'m interested in computer vision and perception of autonomous vehicles. Please contact me through e-mail directly if you are interested, LinkedIn DM is somehow blocked in China mainland so I might not be able to response in time.
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.