FOMO detection result has gradual horizontal position error when deployed to computer

Question/Issue:
FOMO detection result has gradual horizontal position error when deployed to computer

Project ID:
https://studio.edgeimpulse.com/studio/118189/

Context/Use case:
FOMO detection correct on left side:


FOMO detection incorrect on right side:

Thank you in advance :slight_smile:

This issue seems to be specific to the “Computer” deployment target.
The position of the detection corresponds to a 1:1 ratio image, even though the Computer camera is 640x480.

The “Mobile” deployment target is working fine.


Hello @AlbertaBeef,

I cannot reproduce your issue (mostly because I do not have your cards).

However, it recognize several cards when there is obviously none. That’s an accuracy issue, I am not worry that you will be able to collect more backgrounds types to strengthen your model :wink:

Which browser are you using? I also tested the browser inference on a face detection project and I cannot reproduce either.

Let me know

Louis

@Louis,
Just back from vacation, sorry for the delay in my response.
It works fine on Windows, which configures the camera as 1:1 ratio.
I am only seeing the issue when deploying to a Linux (Ubuntu 20.04.4 LTS) computer.
It this case, the camera is configured as 640:480 ratio, and the detection results are displayed on a 480:480 grid within this image.
Hope this helps :slight_smile:
Mario.

This definitely looks like a bug regarding the squashing to square for training/inference, but then not doing the equivalent projection for inference in browser… I’d bet if we scaled those horizontal values by 640/480 it’d line up…

@louis ; does the face detection project you mention work with a square input for in browser inference?

have raised an issue for this one @ AlbertaBeef and will post back when it’s fixed, thanks! Mat

Hey @matkelcey,

Yes, and the collected images are also in a square shape

1 Like