opencv4nodejs

The library provides an asynchronous API. You can consume it in three ways:

1. Callbacks: Pass a callback function as the last argument (e.g., cv.imreadAsync(path, (err, mat) => { ... })).
2. Promises: If no callback is provided, the method returns a Promise.
3. Async/Await: Use await with the Promise-returning methods for cleaner code.

Note: Many methods have an Async suffix (e.g., detectMultiScaleAsync, bgrToGrayAsync).

const cv = require('@u4/opencv4nodejs');

// Using async/await
try {
  const img = await cv.imreadAsync('./faceimg.jpg');
  const grayImg = await img.bgrToGrayAsync();
  const { objects, numDetections } = await classifier.detectMultiScaleAsync(grayImg);
} catch (err) {
  console.error(err);
}

On Windows, you need Windows Build Tools to compile OpenCV and opencv4nodejs. If you do not have Visual Studio or Windows Build Tools installed, you can install the VS2015 build tools globally via npm.

npm install --global windows-build-tools

Install the package using npm. Note that native node modules are built via node-gyp, which requires Python to be installed on your system.

Important: Ensure your project directory path does not contain any whitespaces, as node-gyp may fail to handle them correctly, leading to header file errors (e.g., fatal error C1083: Cannot open include file: 'opencv2/core.hpp').

npm install --save @u4/opencv4nodejs

Once Darknet is compiled, download the pretrained weights and start the training process.

1. Download pretrained weights:

   wget https://pjreddie.com/media/files/darknet53.conv.74 -O darknet53.conv.74

2. **Run Training**:
   Execute the darknet detector training command, providing paths to your `.data` file, `.cfg` file, and the downloaded `.conv.74` weights. Output is redirected to `train.log`:
   ```bash
   ~/darknet-vNext/darknet detector train darknet.data darknet-yolov3.cfg darknet53.conv.74 > train.log

Upon completion, the trained weights will be located at ./weights/darknet-yolov3_final.weights.

By default, installing @u4/opencv4nodejs triggers an auto-build script using the opencv-build npm package. This requires git and a recent version of cmake to be installed on your system.

Customizing the Build

You can control the build process using environment variables:

• Custom Flags: Use OPENCV4NODEJS_AUTOBUILD_FLAGS to pass space-separated CMake flags. This is useful for building only specific modules to reduce compilation time (e.g., export OPENCV4NODEJS_AUTOBUILD_FLAGS=-DBUILD_LIST=dnn).
• Specific Version: Set OPENCV4NODEJS_AUTOBUILD_OPENCV_VERSION to install a specific OpenCV version (e.g., export OPENCV4NODEJS_AUTOBUILD_OPENCV_VERSION=4.6.0).

Follow these steps to prepare your dataset from OpenImages:

1. Fetch Data: Run the script to download and locate relevant OpenImages files and OpenImagesV4 data:

   ts-node getDataFromOpenImages_snowman.ts

2. Split Dataset: Create the train-test split from the JPEGImages directory:

   ts-node splitTrainAndTest.ts JPEGImages

This guide outlines the process of training a custom YOLOv3 object detector specifically for detecting snowmen using OpenImages data and the Darknet framework. The process involves gathering data, splitting it into training/testing sets, compiling Darknet, training the model, and finally running detection using TypeScript.

Prerequisites & Dependencies

First, install the necessary Node.js dependencies and global TypeScript tools:

pnpm install
pnpm install -g ts-node typescript @types/node

(Optional) To suppress warnings when using ts-node, you can set the following environment variable: export NODE_OPTIONS="-r ts-node/register --no-warnings"

To train the model, you must install and compile the darknet-vNext repository.

1. Clone the repository:

   cd ~
   git clone https://github.com/sowson/darknet-vNext
   cd darknet-vNext

2. Configure the Makefile: Edit the first lines of the Makefile to enable or disable specific features (e.g., GPU, CUDNN, OPENCV, OPENMP, DEBUG). Example configuration:

   • GPU=0
   • CUDNN=0
   • OPENCV=0
   • OPENMP=0
   • DEBUG=0

3. Compile:

   make

To build OpenCV automatically using the built-in builder, you can specify the version and optimize the build process.

Speed up builds

Define the OPENCV_BUILD_ROOT environment variable to process the OpenCV build outside of your node_modules folder.

OPENCV_BUILD_ROOT=~/opencv

Specify OpenCV version

You can control which version is built using one of these methods:

1. Environment Variable: Set OPENCV4NODEJS_AUTOBUILD_OPENCV_VERSION="4.9.0".

2. package.json:

"opencv4nodejs": {
  "autoBuildOpencvVersion": "4.6.0"
}

3. build-opencv CLI:

npm link
build-opencv --version 4.6.0 rebuild

To use the module with Electron, you must rebuild it for the Electron runtime. Add a rebuild script to your package.json using the build-opencv command with the --electron flag.

"electron-rebuild": "build-opencv --electron --version 4.5.4 build"

Then run:

npm run electron-rebuild

To avoid building OpenCV inside your container, you can use the pre-built image urielch/opencv-nodejs.

Important: Since the image already has opencv4nodejs installed globally, do not include @u4/opencv4nodejs in your package.json. Instead, use the global installation by setting the NODE_PATH environment variable to /usr/lib/node_modules in your Dockerfile.

FROM urielch/opencv-nodejs