Face-api.js: Using model SSD MobileNet v1 detects only 2 faces max

[hawxxx]

So, I have tested the API in multiple computers. All of them have really good hardware. On my desktop the picture bbt1.jpg will have all 4 faces detected. However, there are 2 computers that no matter if I clone the Face API with no modifications and try with 4 different browsers, the faces detected on the picture bbt1.jpg will only be Raj and Leonard only (2 faces max) this only occurs when using SSD MobileNet v1. If I run the same build on my phone this doesn't occur and all faces are detected. So, it seems that the problem is only on the computers. Any ideas about a similar issue?

[justadudewhohacks]

Can you please give me the exact hardware information of the computers it is not working for (CPU, GPU, OS).

There still seems to be some issues with Intel Graphics Cards / (CPUs?).

Some reference issues: #43 #332 https://github.com/tensorflow/tfjs/issues/510

[hawxxx]

Here are the specs:

CPU: Intel i7-4770
GPU: Intel HD Graphics 4600
OS: Windows 10

Both have the same

[Nk911]

Hi @justadudewhohacks ,

There seems an issue with the tfjs webgl backend, It is being fixed by tfjs team.
But the current version of face-api is not supporting latest tfjs version.
When are you planning to upgrade it?

[justadudewhohacks]

Upgrade to the latest tfjs-core version will come this weekend.

[Nk911]

Hi @justadudewhohacks,
This Issue still persist with the latest version of face-api and tfjs.
Code works properly with 'cpu' backend on tfjs but fails to give proper result on 'webgl' backend on tfjs on the system mentioned above with Intel GPU :

Intel GPU:
Intel HD Graphics 4600 + win 7 ultimate

Any suggestions/solution?