Use Case

Computer Vision

Real-Time Computer Vision systems that detect whether the object-of-interest (person, obstacle, threat) is present constantly poll on device or in the cloud. Running large CNN creates a large AWS cost, and hastens GPU degradation and reduces battery life.

  • Problem: Full model inference shortens battery life
  • With moco: Early exit on frames with obviously no object-of-interest
  • Impact: Reduced compute per frame
  • Outcome: Lower energy consumption → longer battery life