Enhancing 3D Data Processing Efficiency to Propel Next-Generation Algorithm Development of 3D Perception Network
STRADVISION, a leader in deep learning-based vision perception technology, announced the successful deployment of its Data Management Workflow. This advanced system is designed to support the development and mass production of the SVNet 3D Perception Network and was showcased to customers at CES® 2025.
The Data Management Workflow developed by STRADVISION integrates end-to-end automation and optimization, spanning data collection, processing, labeling, and cost settlement. By transitioning these processes to a cloud infrastructure, the workflow achieves significant advancements in data quality, operational efficiency, and cost reduction.
With the growing demand for 3D data processing—critical for autonomous driving and advanced driver assistance systems (ADAS)— STRADVISION has expanded its focus from 2D vision-centered data to 3D data, including LiDAR and multi-camera datasets. This shift necessitated an innovative, automated solution capable of handling large-scale data efficiently while improving data labeling quality and minimizing costs.
Also Read: Arbe Collaborates with NVIDIA to Enhance Radar-Based Free Space Mapping
To meet these challenges, STRADVISION developed its Data Management Workflow, featuring:
- High-quality data collection with SURF: Leveraging the STRADVISION Unified Recording Framework (SURF), this system integrates diverse sensor data to maximize data quality.
- Automated data pipeline: With a scalable architecture based on Kubernetes clusters and Airflow, the data pipeline optimizes computing server resources for high-performance algorithms and automates data processing, enabling efficient handling of large-scale data.
- Web-based labeling tool (Labelit): Supports labeling interfaces for multi-channel cameras and 3D data, enabling the labeling of various ADAS products within a single tool. By leveraging high-quality pre-labeling results from ALT (Auto Labeling Tool) and real-time corrections provided by ALAS (Auto Labeling Assistant Service), it maximizes efficiency, offers accurate workflow guidance, reduces error rates, and enhances data quality.
- Settlement and analysis automation: Employing Workload Logging Service (WLS) and Workload Replay Service (WRS) to automate task analysis and cost management of labeling, driving operational efficiency.
Internally, this innovative workflow enables STRADVISION to accelerate data processing and quality assurance. Externally, it empowers customers by delivering high-quality data quickly, supporting their technology development and performance optimization efforts.
Recent collaborations have highlighted STRADVISION’s expertise in data processing and technological leadership. For example, its Auto Labeling Assistant Service(ALAS) and Deskewing(motion compensation for LiDAR and video data) solutions received high customer praise, reinforcing its position as a trusted partner in the industry.
“This Data Management Workflow represents more than just operational efficiency; it is a critical enabler for the commercialization of autonomous driving technology,” said Junhwan Kim, CEO of STRADVISION. “We are committed to driving innovation and strengthening our global leadership through close collaboration with our customers.”
Source: PRNewswire
