Face-api.js: Issue loading: TypeError: Nt.makeTensor is not a function

[luucv]

I just started and in the latest version I am getting the following error in the browser:

tf-core.esm.js:17 Uncaught (in promise) TypeError: Nt.makeTensor is not a function
    at Sn (tf-core.esm.js:17)
    at kn (tf-core.esm.js:17)
    at o (tf-core.esm.js:17)
    at Fh (tf-core.esm.js:17)
    at tf-core.esm.js:17
    at Array.forEach (<anonymous>)
    at tf-core.esm.js:17
    at Array.forEach (<anonymous>)
    at tf-core.esm.js:17
    at tf-core.esm.js:17

I installed face-api.js with npm install face-api.js. Now using it as follows:

import * as faceapi from 'face-api.js';

await faceapi.nets.tinyFaceDetector.loadFromUri('/models');

As soon as I try to load the model, I get the type error above. Any idea's where it is going wrong? I saw this https://github.com/tensorflow/tfjs/issues/2194#issuecomment-546187719 but couldn't see a solution to that. Thanks!

[hsparrow]

Hi, I just got the same error([email protected]) and I downgraded to [email protected], and this error disappeared, however, a new error rendered. Anyway, you could have a try.

[Spodeopieter]

When downgrading tfjs-core to 1.2.9 I get this error when trying to use the library

SsdMobilenetv1.js:26 Uncaught (in promise) Error: SsdMobilenetv1 - load model before inference
at SsdMobilenetv1.forwardInput (SsdMobilenetv1.js:26)
at SsdMobilenetv1. (SsdMobilenetv1.js:81)
at step (tslib.es6.js:196)
at Object.next (tslib.es6.js:127)
at fulfilled (tslib.es6.js:80)

here is the lock file of a project where this library still works.

yarnlockfile.txt

[luucv]

@Spodeopieter, I think your last error has to do with not asynchronously loading in your model. Are you sure it is being loaded in correctly before you try to use it?

[Spodeopieter]

@luucv this is how I am loading the models :

loadModels = async () => {
const MODEL_URL = "/models";
await faceapi.loadSsdMobilenetv1Model(MODEL_URL);
await faceapi.loadFaceLandmarkModel(MODEL_URL);
await faceapi.loadFaceRecognitionModel(MODEL_URL);
};

[ghost]

@luucv have the same error right now. I installed face-api.js via yarn. I retried it with npm install face-api.js and it worked without any problem.
I'm using webpack-encore with Symfony.

[justlp]

Have the same problem.

Try to delete package-lock.json, node_modules. Then install the latest version (currently 0.21.0)

[justadudewhohacks]

The issue is most certainly due to your package managers installing a tfjs core version other than 1.2.9 (which is the one the latest face-api.js version uses). May be an issue with yarn?

Try deleting your yarn locks / package locks and do a clean installation. If for some reason the issue still remains, try explicitly installing npm i @tensorflow/[email protected].

[ghost]

@justadudewhohacks Thanks for the Tip!
I've solved my problem :D

You do not just have to say "@tensorflow/tfjs-core": "1.2.9" in the dependencies-section of the package.json, you also have to prevent _tfjs-image-recognition-base_ from getting a newer version:

"resolutions": {
    "tfjs-image-recognition-base/@tensorflow/tfjs-core": "<=1.2.9"
}

[giuseppedev]

I've solved by taking these steps:

1. delete tensorflow node module
2. npm i @tensorflow/[email protected]
3. npm i @tensorflow/[email protected]

The 1.2.9 is the last one supported by latest face-api.js version as @justadudewhohacks said.

[sudokzt]

I have same problem.

"@tensorflow/tfjs-core": "1.2.9",
"@tensorflow/tfjs-node": "1.2.9",
"face-api.js": "^0.21.0",

5 days ago, It was operating normally.

[Sxela]

@justadudewhohacks Thanks for the Tip!
I've solved my problem :D

You do not just have to say "@tensorflow/tfjs-core": "1.2.9" in the dependencies-section of the package.json, you also have to prevent _tfjs-image-recognition-base_ from getting a newer version:

"resolutions": {
    "tfjs-image-recognition-base/@tensorflow/tfjs-core": "<=1.2.9"
}

This helped. looks like tfjs-image-recognition-base was secretly installing the latest tfjs behind our backs.

[paulvales]

have same problem, resolutions no helped

[Spodeopieter]

installing face-api with npm instead of yarn seems to solve the problem for me.

[cmaycumber]

installing face-api with npm instead of yarn seems to solve the problem for me.

I'm running into the same issue with yarn, installing with npm solved me problem as well.

[SteveShaffer]

looks like tfjs-image-recognition-base was secretly installing the latest tfjs behind our backs.

I added the resolutions section as suggested above and by this help doc but it was apparently ignored by yarn cuz it resolved to 1.3.1 anyway per yarn.lock. I've got a yarn monorepo so I'm sure there's either some logical reason or some bug as to why that was the case.

I was able to get around it by manually editing the yarn.lock to change the section for "@tensorflow/tfjs-core@^1.2.9": to be identical to the section for "@tensorflow/[email protected]": (no caret). But that seems super hacky and tenuous. If this is working in npm but not yarn that seems like a yarn bug to me.

[naveensky]

I've solved by taking these steps:

1. delete tensorflow node module
2. npm i @tensorflow/[email protected]
3. npm i @tensorflow/[email protected]

The 1.2.9 is the last one supported by latest face-api.js version as @justadudewhohacks said.

This solved the same issue I was having. Hope this helps somebody.

[JaeDukSeo]

So for me nothing worked what did work was this

FOR YARN USERS

1. add
   "resolutions": {
   "tfjs-image-recognition-base/@tensorflow/tfjs-core": "<=1.2.9"
   }

2. in the package add these lines

   "dependencies": {
   "@tensorflow/tfjs-core": "1.2.9",
   "face-api.js": "^0.21.0",
   "next": "9.1.4",
   "react": "16.12.0",
   "react-dom": "16.12.0",
   "react-stickynode": "^2.1.1",
   "styled-components": "^4.4.1",
   "tfjs-image-recognition-base": "^0.6.2"
   }

actually installing tfjs-image-recognition-base worked!

[blazinaj]

I have this same issue. I get the error:

when I try to load any model.

I have the models in public/models and am trying to access them with

 await faceapi.nets.ssdMobilenetv1.loadFromUri('/models') 

I've tried all the fixes mentioned above and I don't think it's an issue with versioning, or npm.

Is there another way to load the models? Thanks!

[Foxyf76]

Also having the same issue, though my error message reads:

engine_1.ENGINE.makeTensor is not a function

Tried some of the above fixes with no luck. Using tfjs 1.2.9 also.

[jjs-stripe]

We need to upgrade to TFJS 1.3.x for other reasons. Any idea what's causing this problem with newer TFJS packages? I don't see makeTensor anywhere in the code here or in tfjs-image-recognition-base
I'm guessing maybe the models need to be regenerated for TF 1.3.x?

[hua1995116]

I've solved by taking these steps:

1. delete tensorflow node module
2. npm i @tensorflow/[email protected]
3. npm i @tensorflow/[email protected]

The 1.2.9 is the last one supported by latest face-api.js version as @justadudewhohacks said.

It solved my problem.

[justadudewhohacks]

I upgraded the tfjs-core version the latest release. Also I decided to move tfjs-image-recognition-base to face-api.js, so yarn should not have any reason to install different versions of tfjs-core anymore unless you or one if your dependencies is explicitly installing another version.

[MadalenaSousa]

I'm getting a "Kt.makeTensor is not a function", more specifically:
Uncaught (in promise) TypeError: Kt.makeTensor is not a function
at Rn (face-api.min.js:2426)
at In (face-api.min.js:2407)
at e (face-api.min.js:11444)
at Vh (face-api.min.js:11445)
at face-api.min.js:12054
at Array.forEach ()
at face-api.min.js:12053
at Array.forEach ()
at face-api.min.js:12047
at face-api.min.js:103

I've tried all these resolutions and none of them worked. Could anyone help please?