I noticed a strange behavior of the FlexiCapture classifier: for several samples, I get different classifier results for the first and for the second recognition. I'd like to get an explanation for it.
One of the reasons can be incorrectly rotated pages in the original file.
The following happens:
- During the first classification, it tries to detect a class on a non-rotated page (as it is in the original file).
- It fails and applies the default class.
- Document Definition is matched and rotation settings are applied (if they are enabled in the Document Definition settings).
- During re-recognitions classification is working with already rotated pages, and classifies them correctly.