Ask Your Question
0

OpenCV and Neural Network / AI Accelerators for Mobile Devices

asked 2018-02-11 06:43:30 -0500

Evren gravatar image

Hello all,

I'm trying to understand if we can expect any performance improvements (as in speed of inference / processing) in projects using OpenCV in mobile platforms that have or will have the existing and upcoming NN / AI accelerators such as the IphoneX's AI chip, Qualcomm's Neural Processing Units on the Snapdragon 835 / 845, PowerVR 2NX, etc. After days of searching online, it is not clear to me if OpenCL and hence OpenCV would be accelerated on such platforms out of the box. The prospect of being able to perform complex CV operations, such as the use of OpenCV's DNN module, taking advantage of any acceleration such chips offer is exciting.

Does anyone have any knowledge on the matter? Is OpenCV able to benefit from these chips via OpenCL acceleration?

I am more than happy to read more on the topic, I may have missed some online material out there. If you are aware of any resources, I would be grateful if you can point me to them.

edit retag flag offensive close merge delete

Comments

You can use intel's movidus.but just to implement DNNs.i heard it supports opencv

Shivanshu gravatar imageShivanshu ( 2018-02-11 11:51:42 -0500 )edit

1 answer

Sort by ยป oldest newest most voted
2

answered 2018-02-11 19:30:28 -0500

Tetragramm gravatar image

The answer is, at the moment, not directly. OpenCV will not run faster when you plug in or enable an AI accelerator. If someone comes along and writes a module/backend, then we'd probably be happy to accept it into the DNN module.

You can however, definitely use OpenCV with the accelerators. Take a look at the example at the bottom of this page for the Movidius. You can see it integrated with an OpenCV program. You should be able to do something similar for just about any neural network accelerator.

edit flag offensive delete link more
Login/Signup to Answer

Question Tools

2 followers

Stats

Asked: 2018-02-11 06:43:30 -0500

Seen: 184 times

Last updated: Feb 11 '18