AWS re:Invent 2025 - StradVision's Vision AI journey with AWS (GBL203)

AWS re:Invent 2025 - StradVision's Vision AI Journey with AWS

Industry Trends in Automotive AI

• The automotive industry is facing three key trends:
  1. Massive Sensor Data: Modern cars are equipped with dozens or hundreds of sensors, generating petabytes of data that need to be processed.
  2. Hyperscale Connectivity: Autonomous vehicles require continuous communication with surrounding infrastructure like traffic lights and road conditions.
  3. Accelerated Development: Intense competition is driving the need for faster development cycles and quicker time-to-market.

StradVision's Vision AI Solutions

• StradVision provides vision AI services for autonomous driving, with three key solutions:
  1. Front Vision: Recognizes objects in front of the vehicle and helps avoid or control them.
  2. Surround Vision: Recognizes the entire surrounding space to assist with safe parking.
  3. Multi-Vision: Recognizes obstacles at various distances to help with both parking and driving.

StradVision's Data Flow Architecture

1. Data Ingestion:

   • Challenge: Physically shipping global data to Korea caused delays and security risks.
   • Solution: Leveraged AWS Direct Connect to securely transfer data online, reducing delivery time from weeks to 24 hours.
   • Benefit: Improved productivity and security by moving to cloud-based data archiving on AWS S3 Glacier.

2. Data Processing:

   • Challenge: On-premises computing resources were insufficient and inflexible to handle the heavy data processing requirements.
   • Solution: Deployed their data processing pipeline to the cloud using AWS services like EC2, enabling scalability and flexibility.
   • Benefit: Achieved a 7x increase in productivity and 2x improvement in scalability compared to their previous on-premises setup.

3. Data Labeling:

   • Challenge: Delivering large processed data volumes to global labeling workers caused latency and productivity issues.
   • Solution: Leveraged AWS CloudFront to cache the data and enable low-latency access for labeling workers worldwide.
   • Benefit: Reduced labeling latency by over 90%, boosting worker productivity and data quality.

Advancing with Synthetic Data

• StradVision is integrating synthetic data generation, evolution, and validation into their data flow pipeline.
• This allows them to augment real-world data with synthetic examples, especially for rare or difficult-to-capture scenarios.
• The company is leveraging powerful AWS GPU instances like the M6i.H200 to accelerate the development of these advanced AI technologies.

Business Impact

• StradVision has mass-produced its SVNet vision AI solution in over 4 million vehicles across 35 models globally.
• By optimizing their data flow on AWS, the company has achieved:
  • 7x improvement in productivity
  • 2x increase in scalability
  • Over 50% reduction in overall unit costs
• These innovations have contributed to the continued growth and performance of StradVision's SVNet 3 solution.