Need help with simple face detection and memory management

[NateFaulkenberry]

Hi, I'm trying face-api for the first time, and while it detects faces successfully, I am running into a wall with memory management. I've looked through your examples and tried a bunch of different solutions, but for some reason my tensors don't seem to be disposed, and memory on my server gets eaten up quickly. Just by looking at my code could you point out where I might be going wrong? Also is this library meant to work with Typescript? I appreciate any help. Thank you!

import path from 'path'
import '@tensorflow/tfjs-node'
import * as faceapi from '@vladmandic/face-api'

(async () => {
    const models = path.resolve(process.cwd(), 'models')
    await faceapi.nets.tinyFaceDetector.loadFromDisk(models)
})()

// https://github.com/vladmandic/face-api
const detect = async (buffer) => {
    // @ts-ignore
    const tensor = faceapi.tf.node.decodeImage(buffer)

    const options = new faceapi.TinyFaceDetectorOptions()

    // @ts-ignore
    const faces = await faceapi.detectAllFaces(tensor, options)
    
    // @ts-ignore
    faceapi.tf.dispose(tensor)
    
    // @ts-ignore
    console.log(faceapi.tf.memory())

    return faces.map(face => {
        return {
            x: face.box.x,
            y: face.box.y,
            width: face.box.width,
            height: face.box.height,
            weight: 1.0
        }
    })
}

export { detect }

[vladmandic]

i just run your code in 1k loop, i don't see a tensor leak - can you show what your console.log output is?

but you do have one memory issue - you create new options object on each detect call - why? create options object just once at the time of model load.