Azure AI Document Intelligence
An Azure service that turns documents into usable data. Previously known as Azure Form Recognizer.
1,535 questions
Azure Document Intelligence API Version 2023/07/31 seems to be more buggy than 2022/08/31
![](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(92.8, 21%, 18%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3EI%3C/text%3E%3C/svg%3E)
Inubot
0
Reputation points
It seems that the latest API version of Azure Document Intelligence introduces issues that weren't present with the previous stable, 2022/08/31. After switching one of our implementations over to the new API version, 2023/07/31, we noticed changes regarding word and table detection.
Words. The next screenshots show that an extra word is picked up when using the latest API version, not present when using the old API:
Tables. It seems that extra rows are being picked up with the latest API, not present when using the old API:
We have used the same training docs to train the model using both API versions. I know MS recommends that Custom Neural should be used, but we are using the Layout template.
Has anyone else experienced the same after switching over to the latest API version?
{count} votes
-
romungi-MSFT 43,696 Reputation points Microsoft Employee2023-08-24T13:38:03.02+00:00 @Inubot Is it possible to share the complete document in this case? If you cannot share the document through this platform, you can use the smiley icon from the studio to share this feedback. Since pre-built models are trained on several documents the model that is published behind the API might change with every iteration as new documents are added to train. It would help the FR team if you could share the documents format with dummy data which shows similar behavior. Thanks!!
Sign in to comment