No face detection in Windows 10 Chrome when the app is built with typescript about human HOT 12 CLOSED

Apologies, I was testing out different versions to see if it would fix the issue. When I am back in the office on Monday I will get new screenshots from one of our Windows 10 users with the same version

from human.

What version of typescript does the browser bundle use? Recently had a pr to fix a build issue related to typescript 5.3 and typedoc, had to downgrade to typescript 5.1 for build to work.
EDIT: update wrong typescript version (previously 2.x, should be 5.x)

from human.

it would be helpfull to see inspector screenshots as above for both TS and JS builds so we can compare them and possibly spot the difference. note that i cannot go an recreate your entire project for purpose of troubleshooting.

also, are you using skipLibCheck=true? its highly recommended for TS projects so your build does NOT try to recompile already compiled libraries.

from human.

@vladmandic Yes skipLibCheck is set to true for us. I'm not exactly sure what you wanted me to screen shot with the inspector, can you please give me an example? I've created gists for the bundles from running the apps in ngrok in case that is helpful.

JS
https://gist.github.com/plmok61/46fc91891666f0a9c10551684a524107

TS
https://gist.github.com/plmok61/8f390264e46341e20741d78a8531aa15

@augchan42 we have typescript 4.9.5 in the simplified app,

from human.

Screenshot exactly like you've posted, but for both TS app and JS app.

from human.

Oh sorry I misunderstood. I though you meant screenshots of the actual built JS which seemed kind of odd.

Here are the JS screenshots:

And here are the TS screenshot

from human.

first of all, TS and JS projects are using different versions of Human - there is no point of comparing if we're not even looking at the same thing.

from human.

Great catch on the versions. Typescript is not the issue after all.

It looks like version 3.1.2 is the culprit. We had ^3.0.6 in our package.json. When we installed the exact version of 3.0.7 in our staging app, face detection in Chrome Windows 10 once again works. We then installed 3.1.2 and had no faces detected.

We also noticed that 3.2.0 was just release so we tested it in staging as well and encountered the same no faces detected issue.

Are there any important improvements/fixes that we are missing out on by using 3.0.7? It seems to serve our purposes well.

from human.

versions 3.1 and 3.2 are primarily to support latest versions of typescript, node and tfjs, so no explicit need to upgrade.
if 3.0 works and 3.1 doesn't it could be due to backend math precision and/or optimizations in newer tfjs.
according to screenshots, you're running using webgpu backend, could you try as a test setting backend to webgl and see if that changes the results?

from human.

We just tested 3.1.2 with backend set to webgl. We get some typescript errors from the config but face detection now works in windows 10 chrome.

This is our config.

{
  backend: 'webgl',
  gesture: {
    enabled: false,
  },
  hand: {
    enabled: false,
  },
  body: {
    enabled: false,
  },
  segmentation: {
    enabled: false,
  },
  face: {
    antispoof: {
      enabled: true,
    },
    liveness: {
      enabled: true,
    },
    iris: {
      enabled: true,
    },
    emotion: {
      enabled: true,
    },
    mesh: {
      enabled: true,
    },
    detector: {
      rotation: true,
    },
  },
}

We will continue to use 3.0.7 for now but it is nice to know that we are able to upgrade if we need to. Thank you for all your help!

from human.

webgpu is still work-in-progress, i'm surprised that chrome enabled it by default in latest versions, i would have left it for beta/dev for a while longer. math optimizations are changing weekly and likely some math results are different on 3 or 4th level behind decimal point, but enough to throw the model off.

anyhow, i'll close the issue for now, but feel free to post here with any further questions.

from human.

I recently encountered a similar problem while developing a simple test website.

On one of my computers, whenever I upload an image for human face detection, it always returns an empty array (No face detection).

This issue persists regardless of which web browser I use.
I have tested it with versions 3.0.7, 3.1.2, and the latest version 3.2.0 of the software, and the problem remains consistent.

However, the same website functions correctly on other computers and mobile devices.
So, I'm also puzzled why the same code and the same image (a frontal face photo) produce completely different results.

I have also tested two demo websites provided by the author on the problematic computer:

Demo Website 1: The issue of 'No face detection' consistently occurs on this URL.
Demo Website 2: However, on this URL, the application runs smoothly

The solution that worked for me was modifying the backend setting to 'wasm'.
backend: 'wasm'
After making this change, the face detection started working as expected.

from human.