Microsoft

Charles Chen

My feedback

  1. 1 vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)

    We’ll send you updates on this idea

    2 comments  ·  Form Recognizer  ·  Flag idea as inappropriate…  ·  Admin →
    An error occurred while saving the comment
    Charles Chen commented  · 

    Sean, I think the downside to saving a few bytes is that every consumer has to implement this logic to make use of the bounding box coordinates. While this is not difficult, I think there's a design question of whether it's worth saving a few bytes vs making an easier to use API layer that reduces redundant code for the consumer.

    As an example, it would be even more efficient to include a tokens dictionary as there are sure to be many repetitive tokens ("the", "I", "he", "she", "a", etc.) and instead of referring to the full string, refer to an index in the token table. However, the savings in byte count with this approach results in a less usable API because every consumer will now need to perform a token lookup to make use of the results.

    Charles Chen shared this idea  · 
  2. 1 vote
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)

    We’ll send you updates on this idea

    0 comments  ·  Form Recognizer  ·  Flag idea as inappropriate…  ·  Admin →
    Charles Chen shared this idea  · 
  3. 2 votes
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)

    We’ll send you updates on this idea

    0 comments  ·  Form Recognizer  ·  Flag idea as inappropriate…  ·  Admin →
    Charles Chen shared this idea  · 
  4. 11 votes
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)

    We’ll send you updates on this idea

    6 comments  ·  Form Recognizer  ·  Flag idea as inappropriate…  ·  Admin →
    An error occurred while saving the comment
    Charles Chen commented  · 

    For this flow, you need to use the analyze layout endpoint to kind of perform a pre-analysis by extracting the title and other metadata about the form.

    Then you can use that information to select a model to use.

    What would be nice is to be able to "chain" these two together automatically like associating key words with the model and then have a service endpoint that accepts the document, extracts the potential title lines (using the largest text closest to the top of the page), and then matching that to a model automatically.

    If you have not tried the layouts service yet, you can find the documentation here: https://docs.microsoft.com/en-us/azure/cognitive-services/form-recognizer/quickstarts/python-layout

    I made a short tutorial video on how to use this in C#: https://www.youtube.com/watch?v=PMjIYGPj6gI

  5. 5 votes
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)

    We’ll send you updates on this idea

    0 comments  ·  Form Recognizer  ·  Flag idea as inappropriate…  ·  Admin →
    Charles Chen supported this idea  · 
  6. 8 votes
    Sign in
    (thinking…)
    Sign in with: Facebook Google
    Signed in as (Sign out)

    We’ll send you updates on this idea

    1 comment  ·  Form Recognizer  ·  Flag idea as inappropriate…  ·  Admin →
    Charles Chen supported this idea  · 

Feedback and Knowledge Base