Domain Specific Approximation for Object Detection

TL;DR

This paper explores domain-specific approximation techniques to significantly accelerate object detection in autonomous systems, achieving up to 7.5x speedup without losing accuracy, through static and dynamic image scaling methods.

Contribution

It introduces novel domain-specific approximation methods for object detection and demonstrates their effectiveness in improving speed while maintaining accuracy.

Findings

Up to 7.5x speedup with dynamic approximation

Domain-specific scaling preserves accuracy

AutoFocus runtime exploits dynamic approximation

Abstract

There is growing interest in object detection in advanced driver assistance systems and autonomous robots and vehicles. To enable such innovative systems, we need faster object detection. In this work, we investigate the trade-off between accuracy and speed with domain-specific approximations, i.e. category-aware image size scaling and proposals scaling, for two state-of-the-art deep learning-based object detection meta-architectures. We study the effectiveness of applying approximation both statically and dynamically to understand the potential and the applicability of them. By conducting experiments on the ImageNet VID dataset, we show that domain-specific approximation has great potential to improve the speed of the system without deteriorating the accuracy of object detectors, i.e. up to 7.5x speedup for dynamic domain-specific approximation. To this end, we present our insights toward harvesting domain-specific approximation as well as devise a proof-of-concept runtime, AutoFocus, that exploits dynamic domain-specific approximation.