StradVision, whose AI-based camera perception software is a leading innovator in Advanced Driver Assistance Systems (ADAS) and Autonomous Vehicles, is collaborating with the South Korean social enterprise Testworks to improve the data processing efficiency and accuracy of its road safety technology such as its pioneering SVNet deep learning-based software.
One of SVNet’s key features is to quickly and accurately identify hazardous and potential road conditions using its deep-learning algorithm. In order to train SVNet’s deep neural network by generating annotated data, StradVision uses its own Auto Labeling Tool to automatically detect and label 94% of objects at eight times the speed of a human labeler. The remaining 6% requires human intervention, which is where StradVision and Testworks’ partnership comes in.
“I was looking for a company to entrust with manually labeling StradVision’s detail-sensitive data, and Testworks immediately came to mind because of their stellar reputation with data processing. We support Testworks’ social mission and the quality of their work is world-class, so it is an easy decision for us to develop this partnership,” said StradVision CEO Junhwan Kim.
Empowering developmentally disabled individuals
StradVision works with Testworks, a top data processing company in South Korea, on data labeling while also providing inclusive employment opportunities for developmentally disabled individuals as well as those who may have difficulty finding work after a hiatus, such as retirees and mothers returning to the workplace.
StradVision and Testworks began “beta testing” their collaboration in 2017, shortly after Testworks ended a successful Autism@Work internship program.
“When I received the phone call from StradVision, I got goosebumps at the great timing,” said Testworks CEO Yoon Seok-won, who realized he could find full-time employment for some of Autism@Work’s successful interns. “It was an exciting opportunity to expand Testworks’ business from software testing to the new area of data processing for AI learning.”
Learning from experience
Testworks refined the operational procedures through trial and error as they were met with challenges over the course of the “beta” period. They introduced layers of data reviews and the addition of a project manager to the team, who could help communicate with full-time developmentally disabled employees to ensure the integrity of data processed for StradVision.
“Autistic individuals tend to have unique sensitivity that allows them to notice small things that non-autistic people would simply overlook, including minute errors in StradVision’s data,” said Yoon.
He found that the project manager, a mother who returned to the workplace, was particularly patient with her colleagues and ensured their work ran smoothly – which was immensely helpful in setting up a positive and accepting work environment.
Combined with the team’s attention to detail, this eventually improved the overall data quality to StradVision’s satisfaction.
Accelerating AV technological advancement
Testworks’ team now processes data from StradVision, correcting any labeling errors and adding important information – ensuring that their AI-based software learns from its mistakes and improves over time.
The project allows developmentally disabled individuals to make a significant contribution to the advancement of AV technology by pushing AI-based software to self-correct after the initial stage of human labeling. This significantly speeds up the growth of deep-learning advanced machine algorithms.