Intel® Movidius™ Neural Compute SDK

View the Project on GitHub

Introduction

The Intel® Movidius™ Neural Compute SDK (Intel® Movidius™ NCSDK) enables rapid prototyping and deployment of deep neural networks (DNNs) on compatible neural compute devices like the Intel® Movidius™ Neural Compute Stick. The NCSDK includes a set of software tools to compile, profile, and validate (check) DNNs as well as the Neural Compute API (NCAPI) for application development in C/C++ or Python.

The NCSDK has two general usages:

The following diagram shows the typical workflow for development with the NCSDK:

Note that network training phase does not utilize the NCSDK.

Frameworks

The NCSDK currently supports two deep learning frameworks:

  1. Caffe from Berkeley Vision Labs
  2. TensorFlow™ from Google

See here for more directions for using networks from these frameworks with the NCSDK.

NCSDK Tools

NCAPI

Applications for performing inferences with the NCAPI can be developed either in C/C++ or Python. The NCAPI provides a software interface to communicate with neural compute devices, load compiled network files, manage input/output, and perform inferences.

Installation

View the NCSDK installation instructions or choose an alternate installation method:

Base Examples

There are several example applications and networks included with the NCSDK:

Each example contains a Makefile. Running make help in the example’s base directory will give possible make targets. Running make examples in the base ncsdk directory will build all examples.

Neural Compute App Zoo

Many more examples, including user-submitted networks and applications, can be found at our Neural Compute App Zoo GitHub repository.

Release Notes

See the latest Release Notes.

Intel® Movidius™ User Forum

There is an active user forum for users of the NCSDK to discuss ideas and seek troubleshooting help at https://ncsforum.movidius.com.