English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

In surveillance and search and rescue applications, it is important toperform multi-target tracking (MOT) in real-time on low-end devices. Today'sMOT solutions employ deep neural networks, which tend to have high computationcomplexity. Recognizing the effects of frame sizes on tracking performance, wepropose DeepScale, a model agnostic frame size selection approach that operateson top of existing fully convolutional network-based trackers to acceleratetracking throughput. In the training stage, we incorporate detectability scoresinto a one-shot tracker architecture so that DeepScale can learn representationestimations for different frame sizes in a self-supervised manner. Duringinference, it can adapt frame sizes according to the complexity of visualcontents based on user-controlled parameters. To leverage computation resourceson edge servers, we propose two computation partition schemes tailored for MOT,namely, edge server only with adaptive frame-size transmission and edgeserver-assisted tracking. Extensive experiments and benchmark tests on MOTdatasets demonstrate the effectiveness and flexibility of DeepScale. Comparedto a state-of-the-art tracker, DeepScale++, a variant of DeepScale achieves1.57X accelerated with only moderate degradation ~2.3\ in tracking accuracy onthe MOT15 dataset in one configuration. We have implemented and evaluatedDeepScale++ and the proposed computation partition schemes on a small-scaletestbed consisting of an NVIDIA Jetson TX2 board and a GPU server. Theexperiments reveal non-trivial trade-offs between tracking performance andlatency compared to server-only or smart camera-only solutions.

DeepScale: Online Frame Size Adaptation for Multi-object Tracking on Smart Cameras and Edge Servers