Training job crash if classes are digits

[torkem]

Hi !
I am working on a use case where the object to detect is a serial number engraved on a part.
To achieve that, I've annotated serial number's digits (from 0 to 9). But, when I start the training, the job crashes.
If I add any letter after the digit (1A instead of 1), the training succeeds.
Thanks for the help

[hadikoub]

Hello,

Is it possible to share a sample annotation? are you saving the numbers as strings or numbers?
Additionally, is it possible to share the error log ?

[torkem]

1854537_1.txt
I attached the annotation file I'm using. You'll notice that I've added an A at each digits to make it working.
(I had to rename the file to txt, xml aren't supported by github)

[hadikoub]

I tried to reproduce the error based on the sample file you provided and the error is happening in the TF record generation step.
I pushed a patch to enforce conversion to str for the class name as in your case the numbers are being cast into int and not str which is causing the error.
commit: 6321860
Can you please check if the newer version fixes your issue? Please make sure to include double quotes in the objectclasses.json for class names as well example:

[
    {
        "Id": 0,
        "Name": "1"
    },
    {
        "Id": 1,
        "Name": "2"
    }
]

[torkem]

Sorry for the delay, I confirm that this issue is fixed, thanks @hadikoub !