MobiSR: Efficient On-Device Super-Resolution through Heterogeneous Mobile Processors

TL;DR

MobiSR is a framework that enables efficient on-device super-resolution by intelligently selecting and scheduling models on heterogeneous mobile processors, significantly improving speed while maintaining image quality.

Contribution

The paper introduces MobiSR, a novel framework that optimizes super-resolution on mobile devices through model compression and dynamic scheduling based on image patch difficulty.

Findings

Achieves 2.13x speedup over difficulty-unaware mappings

Achieves 4.79x speedup over single engine implementations

Maintains high image quality with reduced latency

Abstract

In recent years, convolutional networks have demonstrated unprecedented performance in the image restoration task of super-resolution (SR). SR entails the upscaling of a single low-resolution image in order to meet application-specific image quality demands and plays a key role in mobile devices. To comply with privacy regulations and reduce the overhead of cloud computing, executing SR models locally on-device constitutes a key alternative approach. Nevertheless, the excessive compute and memory requirements of SR workloads pose a challenge in mapping SR networks on resource-constrained mobile platforms. This work presents MobiSR, a novel framework for performing efficient super-resolution on-device. Given a target mobile platform, the proposed framework considers popular model compression techniques and traverses the design space to reach the highest performing trade-off between image quality and processing speed. At run time, a novel scheduler dispatches incoming image patches to the appropriate model-engine pair based on the patch's estimated upscaling difficulty in order to meet the required image quality with minimum processing latency. Quantitative evaluation shows that the proposed framework yields on-device SR designs that achieve an average speedup of 2.13x over highly-optimized parallel difficulty-unaware mappings and 4.79x over highly-optimized single compute engine implementations.