Bookmark and Share

Arm Demonstration of Applications using CMSIS-NN and the Arm Compute Library

Tim Hartley, Senior Product Manager at Arm, demonstrates the company’s latest embedded vision technologies and products at the March 2018 Embedded Vision Alliance Vision Industry and Technology Forum. Specifically, Hartley demonstrates various example applications built using Arm software development tools.

CMSIS-NN provides a library of low level machine learning primitives targeting Arm Cortex®-M microcontrollers. Hartley shows an image classification demo based on this library, implementing CIFAR-10 to recognize a range of object types. The demo runs entirely on a single Cortex-M7 CPU. CMSIS-NN was open-sourced in January of 2018.

The Compute Library provides low-level machine learning and computer vision primitives targeting Arm Cortex-A CPUs and Arm Mali® GPUs. Arm NN is an inference engine, providing a bridge between existing neural network frameworks like TensorFlow and Caffe and underlying hardware. Together, these libraries make the task of creating ML (machine learning) for platforms with these processor types significantly easier. Hartley shows an image classification demo built using this software. Both libraries are available open-source.

blairhartley's picture
Last seen: 42 weeks 6 days ago
Level 1: Prestidigitator
Joined: 2014-09-12
Points: 1


The demo sources are not currently available, though the keyword spotting example can be easily replicated using the set of commands already on Github.  We are looking at options for where we can upload binaries for demos generally - I will post back once I know more.

Regards, Tim Hartley
Last seen: 19 weeks 3 days ago
Level 1: Prestidigitator
Joined: 2013-10-08
Points: 1

Is the code of those demos available somewhere? The speaker mentions that everything is based on open-source libraries. It would be a good idea to have a GitHub repo or equivalent for the code of those demos.