Skip to content

[formrecognizer] service bug with analyze custom form (without labels) with input containing blank page #11014

@kristapratico

Description

@kristapratico

This issue is to track service side bug (seen on 4/22). This can be reproduced by sending a multi-page input with a blank page and passing include_text_content=True. There is an off-by-one page index error due to the blank page not being returned with the results.

Copy/pasted issue explained to service team:

I trained on a 3 page document in which page # 2 was completely blank. When I analyzed my form, I got back results in pageResults and readResults, but only for pages 1 and 3 (list of length==2). My call to analyze had the include_text_details=True, however, when I attempted to resolve the pointers in the SDK I got an out-of-index error.

I found that elements references in the pageResult for page # 3 were referencing the readResult like this, for example:

"#/readResults/2/lines/30/words/0"

Since readResult was returned as a list of 2 items, indexing into readResults[2] does not exist.

Edit: fix tests that work around this once service deploys version fixing this bug

Metadata

Metadata

Assignees

Type

No type

Projects

No projects

Milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions