How to select the appropriate vocab for a text recognition training

[lfxuan]

Hello! Why does the list (map (vocab. Index, input_string)) report an error valueerror: substring not found when training the recognition model

[charlesmindee]

Hi @lfxuan,

Thanks for reporting this, would you please give me the following:

• the code sample to reproduce the bug
• your environment (result of the collect_env.py script)
• eventually the image you passed to the OCR

Thanks: 🙏

[lfxuan]

Thank you for your answer!

[fg-mindee]

'Morning @lfxuan 👋

As mentioned by Charles, we would need a bit more information to have a comprehensive answer. But considering your error, I'm guessing you're training on a dataset that has characters outside of the vocab you selected 🤔

You can easily whether this is the case by printing the string that causes this error and then checking whether all characters are included in the vocab https://github.com/mindee/doctr/blob/main/doctr/datasets/vocabs.py (the default one on the script is "french") 👍

If this is the case, try to select a more appropriate vocab for your dataset, and if it doesn't exist yet in docTR, we can discuss whether we should extend the range of it 😁

Have a good day!

[lfxuan]

Thank you very much for your answer! The reason for the error is that the tag mistakenly adds a space when writing the JSON file. The second vocab ['french '] has no space character, which has been solved.