We will be using vladmandic-face-api as it is compatible with tfjs 2.0.
Setting up the project
We wil be setting up the project and install some packages needed for this project. Initial setup that we need to use to setup the API to upload image, and navigate files/folders.
npm install express express-fileupload fs path
And this one is the face-api, tfjs that is also needed by the face api and canvas to draw the detected points.
npm install @vladmandic/face-api @tensorflow/tfjs canvas
Setup express API and a file upload endpoint.
const express = require("express");
const fileUpload = require("express-fileupload");
const app = express();
const port = process.env.PORT || 3000;
app.use(fileUpload());
app.post("/upload", async (req, res) => {
const { file } = req.files;
console.log(file);
res.send("Successfile upload");
});
app.listen(port, () => {
console.log("Server started on port" + port);
});
In the code above, I setup the key as file.
const { file } = req.files;
We'll be using postman for testing the API with form data body and file as a key.
Adding FaceAPI
Download the AI models here. You can play around on any models but for this example we will only use ssd mobile net for face detection.
faceapiService.js
This file is where we use the face api, in main() we initialize the face api, tf and locate the models. in image() is where we pass the image data we upload and decode it to a tensorflow object and we pass that object to detect() wherein it will return the result for the image that we uploaded.
const path = require("path");
const tf = require("@tensorflow/tfjs-node");
const faceapi = require("@vladmandic/face-api/dist/face-api.node.js");
const modelPathRoot = "./models";
let optionsSSDMobileNet;
async function image(file) {
const decoded = tf.node.decodeImage(file);
const casted = decoded.toFloat();
const result = casted.expandDims(0);
decoded.dispose();
casted.dispose();
return result;
}
async function detect(tensor) {
const result = await faceapi.detectAllFaces(tensor, optionsSSDMobileNet);
return result;
}
async function main(file) {
console.log("FaceAPI single-process test");
await faceapi.tf.setBackend("tensorflow");
await faceapi.tf.enableProdMode();
await faceapi.tf.ENV.set("DEBUG", false);
await faceapi.tf.ready();
console.log(
`Version: TensorFlow/JS ${faceapi.tf?.version_core} FaceAPI ${
faceapi.version.faceapi
} Backend: ${faceapi.tf?.getBackend()}`
);
console.log("Loading FaceAPI models");
const modelPath = path.join(__dirname, modelPathRoot);
await faceapi.nets.ssdMobilenetv1.loadFromDisk(modelPath);
optionsSSDMobileNet = new faceapi.SsdMobilenetv1Options({
minConfidence: 0.5,
});
const tensor = await image(file);
const result = await detect(tensor);
console.log("Detected faces:", result.length);
tensor.dispose();
return result;
}
module.exports = {
detect: main,
};
Updating the upload endpoint
We'll be updating the endpoint and test if it is working.
app.post("/upload", async (req, res) => {
const { file } = req.files;
const result = await faceApiService.detect(file.data);
res.json({
detectedFaces: result.length,
});
});
There you go, we got it working and detect the faces.
You can use any photos with people, I used this one for this example.
Draw Detections
Now we'll be adding the detected result and draw those into the image to see if it really detects the correct faces.
saveFile.js
We'll create a new utils folder with this file to add a utility to save the detected images. We are also adding a out folder where we keep the detected images
const fs = require("fs");
const path = require("path");
const baseDir = path.resolve(__dirname, "../out");
function saveFile(fileName, buf) {
if (!fs.existsSync(baseDir)) {
fs.mkdirSync(baseDir);
}
fs.writeFileSync(path.resolve(baseDir, fileName), buf);
}
module.exports = {
saveFile,
};
Updating faceapiService.js
We are adding this code to draw the detections
const canvas = require("canvas");
const { Canvas, Image, ImageData } = canvas;
faceapi.env.monkeyPatch({ Canvas, Image, ImageData });
async function main(file, filename){
//...Existing code
const result = await detect(tensor);
console.log("Detected faces:", result.length);
const canvasImg = await canvas.loadImage(file);
const out = await faceapi.createCanvasFromMedia(canvasImg);
faceapi.draw.drawDetections(out, result);
save.saveFile(filename, out.toBuffer("image/jpeg"));
console.log(`done, saved results to ${filename}`);
}
Updating upload endpoint
Now we are almost finish, we are going to expose the saved image and add a url to the upload response.
app.post("/upload", async (req, res) => {
const { file } = req.files;
const result = await faceApiService.detect(file.data, file.name);
res.json({
detectedFaces: result.length,
url: `http://localhost:3000/out/${file.name}`,
});
});
app.use("/out", express.static("out"));
Now we'll try the postman again.
There you go we got the detected faces, able to draw and expose the image.
I also created a github repository for reference. face-api.
Top comments (2)
Another outdated article. This is why I hate AI: there is simply not enough information about it. You either have to develop it on your own or ask someone else to do that for you. All articles are outdated, misleading, has broken code or not enough clear. I have SyntaxError: Unexpected token < in JSON at position 6 error because face-api can't find my manifest. I did everything like in text, but it can't find manifest. I dig into source code of the face api library and it seems to be broken, because JSON.parse on 1672 line gets result of readFile function which read manifest from random disk url which somehow constructed somewhere in library code. Disappointed by code quality :(
I tried this fork, and it worked great! The replacement of the original faceapi library was simple and the installation of the support libraries came with no errors. I am using it in an electronjs context so errors commonly pop up, hehe. It runs around 15-30% faster and loads faster.