stradvision

Seoul, Korea - December 11, 2023 - STRADVISION, an automotive industry pioneer in deep learning-based vision perception technology, announced today that its CTO, Jack Sim, presented a groundbreaking study, "Efficient Deep Learning Development through Teacher-Student Learning and Sprints and Its Application to 3D Perception Systems," at Tech.AD USA 2023. The event, focused on 'level 3 to X autonomy & autonomous vehicle technologies,' took place from December 6 to 8, 2023, at The Henry, Dearborn, Michigan.

In his presentation, Sim unveiled an efficient software development process for multiple deep-learning tasks, leveraging teacher-student learning and sprints. This methodology, concurrently improving accuracy and speed, resulted in a remarkable reduction of product development time by up to 50%.

The process was successfully applied in the creation of STRADVISION's augmented reality product, ImmersiView™. Sim explained that the approach decouples dependencies between tasks, trains individual teacher networks in sprints for improved accuracy, and employs a unified student network for simultaneous task learning.

"To optimize accuracy while maintaining inference speed constraints, STRADVISION applied neural architecture search to the student network in the initial sprints and subsequently fixed the architecture in the remaining sprints," Sim explained. This efficient process enabled the first prototype creation within three months, followed by consistent monthly software releases.

The methodology was further extended to address the challenges of building multi-modal 3D perception systems, demonstrating its adaptability and effectiveness in handling diverse sensor inputs during training. Sim emphasized, "This approach ensures comprehensive and accurate perception, exemplifying the adaptability and effectiveness of our development process in handling diverse sensor inputs in training."

Jack Sim concluded, "Our research underscores the significance of our method in advancing the field of multi-modal perception in autonomous systems, paving the way for the future of automotive technology."

STRADVISION continues to lead the way in deep learning-based vision perception technology, driving innovation and setting new standards for efficient and adaptive development processes.