Low Power Inference for On-Device Visual Recognition with a Quantization-Friendly Solution
This work addresses the need for efficient computer vision on mobile devices, though it is incremental as it builds on MobileNets for a competition.
The paper tackled the problem of low-power on-device visual recognition by proposing a quantization-friendly framework for MobileNets, achieving 72.67% accuracy with 27ms latency on a Google Pixel2 phone, which outperformed existing real-time models.
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.