josslei / obj-detect_demo_2020-2021_cpp-process

VGG16 resolving server written in C++, planned to be a child process of a Node.js main GUI process

Chinese Academy of Science Project Demonstration - Invoking of TorchScript module

This is a program that could give classification prediction by invoking a given trained TorchScript module. Planned to run as an independent process, which will only load the module once.

Now it is fit for Node.js program to load as a child process, communicate with reading/writing its standard output/input.

---

Compile Environment

Library	Architecture	Version
LibTorch	x64 for CPU	1.7.1
jsoncpp	x64	1.9.4

Usage

Create as a child process

var child_process = require('child_process');
var child = child_process.spawn('./invoke.exe', ['./assets/module.pt']); 

Read from stdout of the child process

child.stdout.on('data', function (data) {
    console.log(data.toString());
});

Write to stdin of the child process

child.stdin.write("Msg from Node.js!\n");
child.stdin.write("Msg2 from Node.js!\n");

---

Communicate Protocol

Use Json to communicate. Every thing sended or received is Json format string.

To send a Json string stream to the invoking process, the stream should not contain MORE THAN ONE line break ('\n'). You should have ONE AND ONLY ONE '\n' at the end of the message in order to flush the stream correctly.

The format should be restricted like this:

{
    "type" : "XXX",
    "data" : "XXX"
}

"XXX" means the value you're trying to send.

The value of "data" is not restricted to any specific type. It could also be an array, etc..

Of course, the Json stream should be like this when sending the message:

{ "type" : "XXX", "data" : "XXX" }