Face-api.js: 'multiply' not yet implemented or not found in the registry

[archeelp]

Following are the versions of node-packages to reproduce the error
"face-api.js": "^0.22.2"
"@tensorflow/tfjs-backend-cpu": "^2.7.0"
"@tensorflow/tfjs-backend-webgl": "^2.7.0"
"@tensorflow/tfjs-converter": "^2.7.0"
"@tensorflow/tfjs-core": "^2.7.0"

I tried loading the tinyFaceDetectorModel, here is the code snippet
await faceapi.nets.tinyFaceDetector.loadFromUri('/models');
const input = document.querySelector("img");
const result = await faceapi.detectSingleFace(input,new faceapi.TinyFaceDetectorOptions());

It leads to this error:
backend.js?8b87:504 Uncaught (in promise) Error: 'multiply' not yet implemented or not found in the registry. This kernel may not be supported by the tfjs backend you have chosen

[framp]

This looks like a regression in tensorflowjs, downgrading to 2.6.0 fixes the issue

[archeelp]

This looks like a regression in tensorflowjs, downgrading to 2.6.0 fixes the issue

I downgraded to TF 2.6.0 and now it is giving some new error here it is:
Uncaught (in promise) TypeError: Cannot read property 'length' of undefined
at parseTupleParam (conv_util.js?b818:239)
at tupleValuesAreOne (conv_util.js?b818:386)
at Module.eitherStridesOrDilationsAreOne (conv_util.js?b818:390)
at Object.maxPool [as kernelFunc] (MaxPool.js?898d:27)
at kernelFunc (engine.js?6ae2:425)
at eval (engine.js?6ae2:477)
at Engine.scopedRun (engine.js?6ae2:318)
at Engine.runKernelFunc (engine.js?6ae2:475)
at cl (tf-core.esm.js?953f:17)
at maxPool_ (tf-core.esm.js?953f:17)

@framp I think that downgrading didn't work

[framp]

This works ok on my machine: https://gist.github.com/framp/ac13e105f0d1a4209b12047781321a7a

[shailjaa]

I am getting the excat same issue @archeelp
Did you find any solution ?

[dalelane]

for what it's worth, I also get this error when using TFJS 2.7.0

backend.js:665 Uncaught (in promise) Error: 'multiply' not yet implemented or not found in the registry. This kernel may not be supported by the tfjs backend you have chosen
    at notYetImplemented (backend.js:665)
    at MathBackendWebGL.multiply (backend.js:176)
    at Kt.runKernelFunc.a (face-api.min.js:1)
    at engine.js:625
    at engine.js:433
    at Engine.scopedRun (engine.js:444)
    at Engine.tidy (engine.js:431)
    at kernelFunc (engine.js:625)
    at engine.js:639
    at Engine.scopedRun (engine.js:444)

But trying other 2.x releases (I tried TFJS 2.2.0, 2.3.0, 2.4.0, 2.5.0, 2.6.0) all threw this error

face-api.min.js:1 Uncaught (in promise) TypeError: t.batchNormalization is not a function
    at Kt.runKernelFunc.x (face-api.min.js:1)
    at engine.js:625
    at engine.js:433
    at Engine.scopedRun (engine.js:444)
    at Engine.tidy (engine.js:431)
    at kernelFunc (engine.js:625)
    at engine.js:639
    at Engine.scopedRun (engine.js:444)
    at Engine.runKernelFunc (engine.js:636)
    at yu (face-api.min.js:1)

_Update_: I've just seen that this was already reported/known

[shailjaa]

@dalelane I used TFJS 1.9 and it solved that error.

[archeelp]

@shailjaa downgrading to TFJS 1.* will be a workaround.
But if you have other packages which have dependency requirements of TFJS version >= V2.0 it will again cause a series of errors.

[vladmandic]

it's well known that face-api is not compatible with tfjs 2.0+ - you just run into few issues (such as batchNormalization being deprecated so models need updating, not just JS code, etc.)

i've spend a lot of time to rewrite parts, if you wish you can check out https://github.com/vladmandic/face-api