Hello,
I am currently using a custom trained model to extract fields from PDF invoices. Overall, its performance is average, but in certain cases, it mistakenly detects the invoice number.

Even when I manually drag a rectangle around the number, the model still includes the preceding ":" symbol, resulting in an incorrect detection.
Is there a way to instruct the model that the correct number format should be "104263155" without the ":"?
I'm concerned that if I retrain the model using this specific document, it might worsen its performance. By doing so, I fear that the model will be reinforced to recognize ":104263155" as the correct value, leading to further prediction failures.
Thank you in advance for your assistance.