Development of Real-time ADAS Object Detector for Deployment on CPU

Abstract

In this work, we outline the set of problems, which any neural network for object detection faces when its development comes to the deployment stage and propose methods to deal with such difficulties. We show that these practices allow one to get neural network for object detection, which can recognize two classes: vehicles and pedestrians and achieves more than 60 frames per second inference speed on Core\(^{\mathrm{TM}}\) i5-6500 CPU. The proposed model is built on top of the popular Single Shot MultiBox Object Detection framework but with substantial improvements, which were inspired by the discovered problems. The network has just 1.96 GMAC (GMAC – billions of multiply-accumulate operations) complexity and less than 7 MB model size. It is publicly available as a part of Intel® OpenVINO\(^{\mathrm{TM}}\) Toolkit.

Original document

The different versions of the original document can be found in:

• http://arxiv.org/abs/1811.05894

• http://arxiv.org/pdf/1811.05894

• http://link.springer.com/content/pdf/10.1007/978-3-030-29516-5_55,

http://dx.doi.org/10.1007/978-3-030-29516-5_55 under the license http://www.springer.com/tdm

• https://dblp.uni-trier.de/db/journals/corr/corr1811.html#abs-1811-05894,

https://arxiv.org/abs/1811.05894,

https://link.springer.com/chapter/10.1007%2F978-3-030-29516-5_55,

http://arxiv.org/pdf/1811.05894.pdf,

https://academic.microsoft.com/#/detail/2969890494