Turbo: Opportunistic Enhancement for Edge Video Analytics

TL;DR

This paper introduces Turbo, a system that opportunistically utilizes idle GPU resources at the edge to enhance video data quality, significantly improving object detection accuracy without increasing latency.

Contribution

Turbo presents a novel approach for leveraging non-deterministic idle GPU resources for data enhancement in edge video analytics, combining task-specific modules, adversarial training, and resource-aware scheduling.

Findings

Boosts object detection accuracy by 7.3-11.3%

No additional latency incurred during enhancement

Effective across multiple pipelines and datasets

Abstract

Edge computing is being widely used for video analytics. To alleviate the inherent tension between accuracy and cost, various video analytics pipelines have been proposed to optimize the usage of GPU on edge nodes. Nonetheless, we find that GPU compute resources provisioned for edge nodes are commonly under-utilized due to video content variations, subsampling and filtering at different places of a pipeline. As opposed to model and pipeline optimization, in this work, we study the problem of opportunistic data enhancement using the non-deterministic and fragmented idle GPU resources. In specific, we propose a task-specific discrimination and enhancement module and a model-aware adversarial training mechanism, providing a way to identify and transform low-quality images that are specific to a video pipeline in an accurate and efficient manner. A multi-exit model structure and a resource-aware scheduler is further developed to make online enhancement decisions and fine-grained inference execution under latency and GPU resource constraints. Experiments across multiple video analytics pipelines and datasets reveal that by judiciously allocating a small amount of idle resources on frames that tend to yield greater marginal benefits from enhancement, our system boosts DNN object detection accuracy by $7.3-11.3\%$ without incurring any latency costs.