Azure AI Document Intelligence
An Azure service that turns documents into usable data. Previously known as Azure Form Recognizer.
1,528 questions
prebuilt-layout page orientation marks points inconsistent
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(214.4, 81%, 30%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3EMV%3C/text%3E%3C/svg%3E)
Mauricio Villegas
0
Reputation points
I have been trying out analyze_document with model prebuilt-layout (2024-02-29-preview). I have noticed that the points of the polygons for lines and words encode the text orientation. That is, when looking at the text horizontally (rotating the image if needed) the points always start with the top-left point and the rest of the points are in clockwise order. I think this is great and very convenient! I am not sure if this is documented somewhere. Is this a feature that can be relied on to be preserved?
If this is an intended feature, I observed an inconsistency with the marks (checkboxes). When the image is rotated, the polygon points don't follow the top-left + clockwise order. Is this a bug? Or is there a reason for this inconsistency?
Thank you in advance!
{count} votes
-
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(278.4, 1%, 36%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3EYM%3C/text%3E%3C/svg%3E)
YutongTie-MSFT 48,501 Reputation points2024-05-27T03:12:55.0133333+00:00 Thanks for reaching out to us, we are checking on this internally. A quick question, if every checkbox results in the same issue?
Regards,
Yutong
-
Mauricio Villegas 0 Reputation points
2024-05-27T11:38:12.1566667+00:00 I see it for all checkboxes. Could be related to the fact that checkboxes are not part of text lines? So the checkboxes are not aware of the reading order?
I created a few screenshots that illustrate the behavior. For an image that was straight up, text reading order left-to-right, I rotated it 90, 180 and -90 degrees. Then ran prebuilt-layout for the four rotations. In the screenshots there is a small triangle over the first point in the polygon, and showing the direction to the second point. It can be seen that for the numbers "5" and "6" which are text lines / words, the first point is always top-right with respect to the text. But for the checkboxes, for each rotation the first point is different.
- 0 degrees
- 90 degrees
- 180 degrees
- -90 degrees
- 0 degrees
-
YutongTie-MSFT 48,501 Reputation points
2024-05-30T04:09:57.96+00:00 Thanks for the information, I have forwarded it to product team for evaluation, will let you know once we get any response.
Regards,
Yutong
Sign in to comment