single9 / darknet.js

A NodeJS wrapper of pjreddie's darknet / yolo.

Darknet.JS

A Node wrapper of pjreddie's open source neural network framework Darknet, using the Foreign Function Interface Library. Read: YOLOv3 in JavaScript.

Prerequisites

• Linux, Mac, Windows (Linux sub-system),
• Node (most versions will work, darknet.js <=1.1.5 only works on node <=8.11.2)
• Build tools (make, gcc, etc.)

Examples

To run the examples, run the following commands:

git clone https://github.com/bennetthardwick/darknet.js.git darknet && cd darknet
npm install
./examples/example

Note: The example weights are quite large, the download might take some time

Installation

Super easy, just install it with npm:

npm install darknet

If you'd like to enable CUDA and/or CUDANN, export the flags DARKNET_BUILD_WITH_GPU=1 for CUDA, and DARKNET_BUILD_WITH_CUDNN=1 for CUDANN, and rebuild:

export DARKNET_BUILD_WITH_GPU=1
export DARKNET_BUILD_WITH_CUDNN=1
npm rebuild darknet

Usage

To create an instance of darknet.js, you need a three things. The trained weights, the configuration file they were trained with and a list of the names of all the classes.

import { Darknet } from 'darknet';

// Init
let darknet = new Darknet({
    weights: './cats.weights',
    config: './cats.cfg',
    names: [ 'dog', 'cat' ]
});

// Detect
console.log(darknet.detect('/image/of/a/dog.jpg'));

In conjuction with opencv4nodejs, Darknet.js can also be used to detect objects inside videos.

const fs = require('fs');
const cv = require('opencv4nodejs');
const { Darknet } = require('darknet');

const darknet = new Darknet({
  weights: 'yolov3.weights',
  config: 'cfg/yolov3.cfg',
  namefile: 'data/coco.names'
});

const cap = new cv.VideoCapture('video.mp4');

let frame;
let index = 0;
do {
  frame = cap.read().cvtColor(cv.COLOR_BGR2RGB);
  console.log('frame', index++); 
  console.log(darknet.detect({
    b: frame.getData(),
    w: frame.cols,
    h: frame.rows,
    c: frame.channels
  }));
} while(!frame.empty);

Example Configuration

You can download pre-trained weights and configuration from pjreddie's website. The latest version (yolov3-tiny) is linked below:

• weights
• config
• names

If you don't want to download that stuff manually, navigate to the examples directory and issue the ./example command. This will download the necessary files and run some detections.

Built-With

• Node FFI
• Ref
• Darknet