Facemesh Model Fails to Load (403 Forbidden Errors)

[cecil2084]

Eye tracking no longer works because the facemesh TensorFlow models fail to load. I've verified that this issue also occurs on WebGazer’s official live demo site.

Steps to Reproduce

1. Open WebGazer’s official demo.
2. Check the browser console after allowing camera access.
3. Observe multiple 403 errors from Google Cloud Storage.

[Zainking]

We're experiencing the same issue and are currently conducting paid experiments using your library. This is a critical blocker for our work.

[jeffhuang]

Thanks for reporting, I've also gotten emails about this. It sounds like it started today. I think it's fixable, but we had a baby recently so it might take a little longer than usual to resolve. Thanks for your patience.

[Zainking]

Thanks for reporting, I've also gotten emails about this. It sounds like it started today. I think it's fixable, but we had a baby recently so it might take a little longer than usual to resolve. Thanks for your patience.

Congratulation！

[johngatchemark]

Thanks for reporting, I've also gotten emails about this. It sounds like it started today. I think it's fixable, but we had a baby recently so it might take a little longer than usual to resolve. Thanks for your patience.

We are also currently conducting experiments using your library so I hope that this will be an easy fix 😭😭😭. Anyway, congratulations on the new baby!

[LoganLxb]

I also encountered the same problem

[AndCo1]

I'm also encountering the same problem and need this to conduct an experiment.

[Jani6453]

Hi!
If anyone would like to solve this reference problem locally, they can do so as follows. At least, this is what worked for me:
The Webgazer code can be restored to a more transparent state as follows:
Unminify JS, CSS, HTML, XML and JSON Code

First, you need to download the files from Kaggle:

MediaPipe | facemesh | Kaggle

MediaPipe | iris | Kaggle

TensorFlow | blazeface | Kaggle

Then place it in a local folder, e.g. models/facemesh/, models/blazeface, models/iris/.
Next, you need to add the references in the following functions (just replace the line beginning with Ck..):

function FP(e) { return Qz(this, void 0, void 0, function () { return eP(this, function (t) { return null != e ? [2, Ck(e)] : [2, Ck("./models/facemesh/model.json")]; // Only string! }); }); }

function _P(e) { return Qz(this, void 0, void 0, function () { return eP(this, function (t) { return null != e ? [2, Ck(e)] : [2, Ck("./models/iris/model.json")]; // Only string }); }); }

function lP(e) { var t = void 0 === e ? {} : e, n = t.maxFaces, s = void 0 === n ? 10 : n, r = t.inputWidth, a = void 0 === r ? 128 : r, i = t.inputHeight, o = void 0 === i ? 128 : i, l = t.iouThreshold, u = void 0 === l ? 0.3 : l, c = t.scoreThreshold, h = void 0 === c ? 0.75 : c, p = t.modelUrl; return tP(this, void 0, void 0, function () { var e; return nP(this, function (t) { switch (t.label) { case 0: return null == p ? [3, 2] : [4, Ck(p)]; case 1: return (e = t.sent()), [3, 4]; case 2: return [ 4, Ck("./models/blazeface/model.json") // Only string ]; case 3: (e = t.sent()), (t.label = 4); case 4: return [2, new oP(e, a, o, s, u, h)]; } }); }); }

I am using a slightly modified version of this in Psychopy JS Builder for an experiment, and it is working fine for me now. I hope this will help anyone who is stuck. Go for it!

[kiruthiga-kameswaran]

This error is fixed, thank you for fixing!