How an OEM used Sama to improve a Multi-Class Object Detection (MCOD) System

GOAL

A large auto OEM was building a Multi-Class Object Detection (MCOD) system, with cameras that captured a 360° view of the road, and needed help scaling annotations for millions of samples for semantic segmentation, object tracking, and object detection.

‍

WORKFLOWS

Sama ramped thousands of agents within a month and:

• Built a quality rubric with a 2-pixel tolerance
• Provided full scene semantic segmentation and MCOD metadata tagging
• Annotated objects using bounding boxes for sequence object tracking and object detection, plus polylines and polygons for key map feature sets

‍

RESULTS

Sama achieved a 98% quality SLA and helped the client:

• Annotate millions of samples at scale
• Significantly improve object tracking and detection
• Improve model accuracy and efficiency

‍