# Low Power Inference for On-Device Visual Recognition with a Quantization-Friendly Solution

## TL;DR

This paper presents a quantization-friendly framework for MobileNets that enables low-power, real-time visual recognition on mobile devices, achieving high accuracy and efficiency without custom hardware.

## Contribution

The paper introduces a novel quantization-friendly approach for MobileNets, improving on existing models for on-device visual recognition in terms of accuracy and latency.

## Key findings

- Achieved 72.67% accuracy on holdout dataset.
- Latency of 27ms on a single CPU core of Google Pixel2.
- Outperforms previous real-time MobileNet models.

## Abstract

The IEEE Low-Power Image Recognition Challenge (LPIRC) is an annual competition started in 2015 that encourages joint hardware and software solutions for computer vision systems with low latency and power. Track 1 of the competition in 2018 focused on the innovation of software solutions with fixed inference engine and hardware. This decision allows participants to submit models online and not worry about building and bringing custom hardware on-site, which attracted a historically large number of submissions. Among the diverse solutions, the winning solution proposed a quantization-friendly framework for MobileNets that achieves an accuracy of 72.67% on the holdout dataset with an average latency of 27ms on a single CPU core of Google Pixel2 phone, which is superior to the best real-time MobileNet models at the time.