Shinobi/plugins/tensorflow

TensorFlow.js

Ubuntu and CentOS only

Go to the Shinobi directory. /home/Shinobi is the default directory.

cd /home/Shinobi/plugins/tensorflow

Copy the config file.

• To run the install script interactively:

sh INSTALL.sh

• To run the install script non-interactively:

sh INSTALL.sh

WITH the appropriate flags

Flag	Purpose
--jetson	Install for Jetson Nano
--arm	Install for ARM Processors (like Raspberry Pi Model 3 B+)
--gpu	Install for GPU support
--dont-create-key	Don't generate random plugin key

Start the plugin.

pm2 start shinobi-tensorflow.js

Doing this will reveal options in the monitor configuration. Shinobi does not need to be restarted when a plugin is initiated or stopped.

Run the plugin as a Host

The main app (Shinobi) will be the client and the plugin will be the host. The purpose of allowing this method is so that you can use one plugin for multiple Shinobi instances. Allowing you to easily manage connections without starting multiple processes.

Edit your plugins configuration file. Set the hostPort to be different than the listening port for camera.js.

nano conf.json

Here is a sample of a Host configuration for the plugin.

• plug is the name of the plugin corresponding in the main configuration file.
• https choose if you want to use SSL or not. Default is false.
• hostPort can be any available port number. Don't make this the same port number as Shinobi. Default is 8082.
• type tells the main application (Shinobi) what kind of plugin it is. In this case it is a detector.

{
  "plug":"Tensorflow",
  "hostPort":8082,
  "key":"Tensorflow123123",
  "mode":"host",
  "type":"detector"
}

Now modify the main configuration file located in the main directory of Shinobi.

nano conf.json

Add the plugins array if you don't already have it. Add the following object inside the array.

  "plugins":[
      {
          "id" : "Tensorflow",
          "https" : false,
          "host" : "localhost",
          "port" : 8082,
          "key" : "Tensorflow123123",
          "mode" : "host",
          "type" : "detector"
      }
  ],

Docker Installation

Install Shinobi Plugin with Docker

Image is based on node:12.22.1-buster-slim.

1. Enter plugin directory. Default Shinobi installation location is /home/Shinobi.

cd /home/Shinobi/plugins/tensorflow

2. Build Image.

docker build --tag shinobi-tensorflow-image:1.0 .

3. Launch the plugin.

• -e ADD_CONFIG='{"key":"123mypluginkey","tfjsBuild":"gpu","host":"172.16.100.238","port":"8080"}' Adds any configuration parameters to the plugin's conf.json file.
• -p '8082:8082/tcp' is an optional flag if you decide to run the plugin in host mode.

docker run -d --gpus all --name='shinobi-tensorflow' -e ADD_CONFIG='{"key":"123mypluginkey","tfjsBuild":"gpu","host":"172.16.100.238","port":"8080"}' shinobi-tensorflow-image:1.0

** Logs **

docker logs /shinobi-tensorflow

** Stop and Remove **

docker stop /shinobi-tensorflow && docker rm /shinobi-tensorflow

Options (Environment Variables)

Option	Description	Default
ADD_CONFIG	The plugin's name.	Tensorflow