Recognition of slightly italic digital numbers sometimes fails.

[Brakelmann]

Describe the bug
Recognition of slightly italic digital numbers sometimes fails.
The recognition of the third number after the comma often fails.

Version
V 11.3.1

Screenshots

Recognition:

Is there any chance to get a better recognition rate?

Thanks in advance for any support.

[caco3]

There are already similar digits in the training data, see https://jomjol.github.io/neural-network-digital-counter-readout/
But they are less "italic".

You could try to rotate it a few degrees anti-clock-wise to compensate the tilting and use different y-values for the ROIs

[Brakelmann]

Thanks for you hint.
I gave it a try with 2,3 and 4 degrees anti-clock- wise.
Unfortunately without success.
It is recognizing the 0 alway as 7.

Any other idea?

Thanks, Brakelmann.

[caco3]

can you show another screenshot?

You also can record your data and send them so we can add them to the training data

[Brakelmann]

Sure attached the screenshot with 3 degrees rotated anti clock wise

And digits with its recognition:

I will try to record the digits and provide the grafics.
Would the last digit be sufficient?

[caco3]

try to tilt it more, so the numbers are vertically.

And maybe you have to make the ROIs smaller as well, see here: https://github.com/jomjol/AI-on-the-edge-device/wiki/ROI-Configuration#4-model-selection

[Brakelmann]

Yipiie I got a first success.
Tilted to 6 degrees and made the ROIs a bit smaller.

Attached the screenshots with success.
I will keep you updated if it is a stable recognition.

Thanks, you saved my weekend.