multiple_device.md

Multiple Devices Support

This project supports multiple devices in parallel for classification/detection. Please make sure you have already set up environment and installed this project correctly.

1 Introduction

Movidius NCS SDK supported multiple NCS in parallel for acceleration. We integrate this feature into this project in v0.6.0 release. You can now accelerate your classification/detection for both image and stream input following below guides.

2 Arguments

Arguments	Description	Default Value	Valid Values
max_device_number	max device number in one node	255	0~N-1(N is the maximum number of inserted availabe NCS devices)
start_device_index	start index of all inserted NCS devices in one node	0	0~N-1(N is the maximum number of inserted availabe NCS devices)

3 Guide

In order to control the inserted devices, we add two arguments as described above. One example of how to use these arguments is in the launch files ncs_camera.launch in line 42 and line 43. By default, one process will use at most 255 devices, and use the start index as 0.
You can specify other choices of these arguments in your own launch files. Make sure they are in the valid scope.

4 Demo

Launch two object detection services on two consoles as comparison.

roslaunch movidius_ncs_launch single_ncs_demo.launch cnn_type:=mobilenetssd
roslaunch movidius_ncs_launch multiple_ncs_demo.launch cnn_type:=mobilenetssd

Run the example application on another console. For example,

roslaunch movidius_ncs_launch ncs_image_parallel_detection_example.launch image_base_path:=<absolute-image-directory>