Leveraging Transprecision Computing for Machine Vision Applications at the Edge

TL;DR

This paper introduces a lightweight, dynamic approach for resource-constrained edge devices to optimize machine vision performance by balancing accuracy, throughput, and energy consumption through transprecision computing.

Contribution

It presents a novel runtime workload-aware optimization method that adaptively switches configurations to maximize QoS in machine vision tasks at the edge.

Findings

Achieves 1.6x higher frame processing rate with only 1% accuracy drop.

Effectively manages accuracy-throughput trade-offs in real-time.

Demonstrates improved energy efficiency at the edge.

Abstract

Machine vision tasks present challenges for resource constrained edge devices, particularly as they execute multiple tasks with variable workloads. A robust approach that can dynamically adapt in runtime while maintaining the maximum quality of service (QoS) within resource constraints, is needed. The paper presents a lightweight approach that monitors the runtime workload constraint and leverages accuracy-throughput trade-off. Optimisation techniques are included which find the configurations for each task for optimal accuracy, energy and memory and manages transparent switching between configurations. For an accuracy drop of 1%, we show a 1.6x higher achieved frame processing rate with further improvements possible at lower accuracy.