Using Immich Face Recognition via API

[Zefau]

The Immich API (https://immich.app/docs/api/) does not provide a possibility to run an external image, e.g. from the front-door camera through face recognition (or am I wrong?).

Is there any way to use the defined faces in Immich as a foundation for a face recognition?
So far I was only able to get face detection applied on an external image.

[bo0tzz]

#9285

[Zefau]

For people coming across this in future: The Immich API does not support face recognition for "external" images (images not known by Immich as an asset).

As a workaround, you may query the postgres database yourself to overcome this and use Immich and all the faces in Immich as your face recognition.

1. Run face detection via HTTP-POST request to http://immich-machine-learning:3003/predict (see payload in example below)
2. Use the embedding returned by the request to query the database looking for faces / persons close to the embedding

EXAMPLE

Remark: Fill in your database connection details as well as URL for the machine learning instance.

/*
 * set sql query
 */
const query = `
    WITH "cte" AS (
        SELECT "person"."name", "face_search"."embedding" <=> '$1' AS "distance"
        FROM "asset_faces"
        INNER JOIN "face_search" ON "face_search"."faceId" = "asset_faces"."id"
        LEFT JOIN "person" ON "person"."id" = "asset_faces"."personId"
    )

    SELECT "name", MIN("distance") AS "distance"
    FROM "cte"
    GROUP BY "name"
    HAVING "name" <> ''
	ORDER BY "distance"
`

/**
 * set face detection options
 */
function face(file) {
    let faces = [];
    
    // connect to db
    const client = new Client({
        'user': '',
        'password': '',
        'host': '192.168.0.0',
        'port': '5432',
        'database': 'immich'
    });
    
    // run face detection
    const data = new FormData();
    data.append('entries', '{ "facial-recognition": { "recognition": { "modelName": "buffalo_l", "options": { "minScore": 0.035 }}, "detection": { "modelName": "buffalo_l" }}}');
    data.append('image', fs.createReadStream(file));

    const options = {
        'method': 'post',
        'maxBodyLength': Infinity,
        'url': 'http://immich-machine-learning:3003/predict',
        'headers': { ...data.getHeaders() },
        data
    };
    
    return axios.request(options)
        .then(result => {
            faces = result?.data?.['facial-recognition'] || [];
            return client.connect();
        })
        .then(() => {
            return Promise.allSettled(
                faces.map(face => {
                    return new Promise((resolve, reject) => {
                        client.query(query.replace('$1', face.embedding), (error, result) => {
                            if (error || !result || !result.rows || !result.rows[0] || !result.rows[0].name) {
                                return reject({ 'name': null, 'score': 0, error });
                            }

                            return resolve({ ...result.rows[0], 'score':  Math.round(face.score * 100) });
                        })
                    })
                })
            );
        })
        .then(result => {
            return { file, 'persons': result.map(res => res.status === 'fulfilled' ? res.value : res.reason) };
        })
        .catch(error => log(error, 'error'))
        .finally(() => client.end());
}

Use like

face('your-image.jpg').then(res => console.log(res));

[uwe72]

i am using CodeProject.AI-Server for Face and carplate detection within JS-Adapter in iobroker. Works very well: