[BUG] MRZ ONE LINE BUG

[Edouard-Tby]

First of all, thank you for the very good job on this project — the model works great in most cases.

However, I encountered an issue where the model fails to correctly process 1-line MRZ (Machine Readable Zone) documents. The attached sample file reproduces the problem consistently.

When submitting a document containing a single-line MRZ, the model either fails to detect the MRZ or returns incorrect parsing results.

No explicit error message is returned, but the MRZ output is invalid.

[zephyr-sh]

Thank you very much for your feedback and for the kind words about the project — we really appreciate you taking the time to report this and to provide a reproducible sample 🙏

Regarding the issue with 1-line MRZ (Machine Readable Zone) documents, I should clarify this upfront:
the model was originally designed and trained with standard 2-line and 3-line MRZ formats in mind. Support for single-line MRZ was not explicitly considered during the initial design phase.

As a result, when processing 1-line MRZ documents, the model may exhibit the following behavior:

• Fail to detect the MRZ region
• Detect the region but produce incorrect parsing results
• Return no explicit error, while the MRZ output itself is invalid

This is not an issue with your input or usage — it reflects a current limitation of the model’s scope (in hindsight, this is one of those edge cases we admittedly overlooked 😅).

The sample you provided is extremely helpful. We will evaluate possible improvements, such as:

• Adding explicit handling or detection logic for 1-line MRZ formats
• Clearly identifying unsupported MRZ layouts at the document level to avoid misleading outputs

For now, if your workflow requires reliable handling of 1-line MRZ, we recommend treating these documents as an unsupported case.

Thank you again for the detailed report. Real-world examples like this are invaluable for improving the project, and any updates or changes will be clearly documented in future release notes.