Vision - Detailed identified breaking changes

[Big Data Mining]

Changes detected in Google Vision that affect document analysis (1).docx

We are too worried about this breaking changes. Its affect our application in production!

The scariest: none of these changes are described in the "releases notes" (https://cloud.google.com/vision/docs/release-notes?hl=pt_br)

[s_g]

@bigdatamin, we have tracked down the root cause and are working on an option to revert back to the previous reading order. We will also update the release note in the future when there are significant AI quality upgrades. Thanks!

[Big Data Mining]

The reading order is one of the big issues. But, what about the others, reported in the document? 

[s_g]

Hi, due to cybersecurity considerations, we would like to avoid opening unknown documents/attachments if possible. 

Do you mind briefly summarizing the "other issues in addition to reading order" here (i.e., in bullet point format)? Thanks!

[Big Data Mining]

• The segmentation of blocks in some cases became less comprehensive, however, the order is very different from the legacy model, as you can see in the example below. 
• Some unordered blocks and the segmentation of some blocks became gener.
  
• Blocks 7 to 14 became a big block in the latest versions.
  
• Reading the seal stamps, before they were just one thing, in the latest versions there are several blocks and they are disordered.
  
• Change in paragraph segmentation and clutter.
  
• Overlapping blocks with a more generic segmentation.
  
• Segmentation of blocks that were previously identified together are now separate. As an example of the joint information at the end of the document.
  
• Unordered blocks.
  

[Big Data Mining]

Just another point: the legacy model will not be avaiable after August 20 and this huge big problems are not solved! I cannot believe that Google is oferring a product ready for production with this kind of versioning issues, bad communcation with the community and not supporting background compatibility.

[s_g]

Hi Team, we would be happy to review your usage and potentially extend the builtin/legacy deadline beyond August 20. Do you have a GCP account representative we can reach out to?

P.S., please refrain from posting duplicate content across multiple threads. It might slow down the issue triage. Thanks! 

[Big Data Mining]

Yes, we have - BDM contact: deric.f...@bdm.net.br Google Account executive contact: condor...@google.com

[s_g]

Thanks! We will reach out to the account executive regarding the legacy model deadline extension. Thanks!

[Big Data Mining]

Hello, we did not receive any update about the extension. Could you help me? We are really worried ab
out that!!

[s_g]

We reached out the account manager - still waiting for a response. 

In the meantime, can you try the following to revert back to the previous behavior? 

Set value to "legacy_layout" string in following function when sending request to Vision API:

https://github.com/googleapis/java-vision/blob/62ce19b72c5cb544819f47b64decc20a9499c6df/proto-google-cloud-vision-v1/src/main/java/com/google/cloud/vision/v1/TextDetectionParams.java#L726

[Big Data Mining]

Let us check it out! Just let me know if the other issues are addressed by this version too:

• The segmentation of blocks in some cases became less comprehensive, however, the order is very different from the legacy model, as you can see in the example below. 
• Some unordered blocks and the segmentation of some blocks became gener.
  
• Blocks 7 to 14 became a big block in the latest versions.
  
• Reading the seal stamps, before they were just one thing, in the latest versions there are several blocks and they are disordered.
  
• Change in paragraph segmentation and clutter.
  
• Overlapping blocks with a more generic segmentation.
  
• Segmentation of blocks that were previously identified together are now separate. As an example of the joint information at the end of the document.
  
• Unordered blocks.

And what about the deadline to keep using this "legacy_layout" version? Do we already have a forecast for the "stable" and "latest" versions to incorporate these fixes as well?

[s_g]

Let's discuss the extension through email. Thanks! 

[sysa...@niptech.net.au]

WE too have the same problem with the reading order of the new model compared with the legacy. This is effecting a production system that has been working for the last 3 years.  We would like to request that the legacy model not be retired until the reading order has been fixed in the new model

[Janaka Kumari]

We too have same problem with reading order.It is affecting the production system which is deployed for multiple clients.Pls retain legacy model until the issue is fixed in new model.

[s_g]

We've just emailed you offline for the extension. Thanks!

[s_g]

We have emailed you offline to extend access to builtin/legacy to mid November. 

In the meantime, please try the following method on builtin/stable to resolve the reading order issue: 

• Try setting the "legacy_layout" value for the advanced_ocr_options field in TextDetectionParams. This is something we have recently implemented in "builtin/stable" OCR version to have a word order similar to "builtin/legacy" while retaining the benefits of more accurate text detection and recognition. 

[Janaka Kumari]

Hi,

I have added the options in features as below where we add the model.

img_requests.append({
                    'image': {'content': ctxt},
                    'features': [{
                        'type': 'DOCUMENT_TEXT_DETECTION',
                        'maxResults': 1,
                        'advanced_ocr_options':'legacy_layout',
                        'model':'builtin/stable',
                       
                    }]

if i change the model to builtin/legacy  and remove the advanced_ocr_options it is working fine.Am i adding the option in wrong place?

I am getting the below error if i try to add the advanced_ocr_options in features:

{
  "error": {
    "code": 400,
    "message": "Invalid JSON payload received. Unknown name \"advanced_ocr_options\" at 'requests[0].features[0]': Cannot find field.",
    "status": "INVALID_ARGUMENT",
    "details": [
      {
        "@type": "type.googleapis.com/google.rpc.BadRequest",
        "fieldViolations": [
          {
            "field": "requests[0].features[0]",
            "description": "Invalid JSON payload received. Unknown name \"advanced_ocr_options\" at 'requests[0].features[0]': Cannot find field."
          }
        ]
      }
    ]
  }
}

with regards,

Jana

--
© 2018 Google Inc. 1600 Amphitheatre Parkway, Mountain View, CA 94043
 
Email preferences: You received this email because you signed up for the Google Cloud Vision Discussion Google Group (cloud-visi...@googlegroups.com) to participate in discussions with other members of the Google Cloud Vision community and the Google Cloud Vision Team.
---
You received this message because you are subscribed to a topic in the Google Groups "cloud-vision-discuss" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/cloud-vision-discuss/pEntAA3iPdI/unsubscribe.
To unsubscribe from this group and all its topics, send an email to cloud-vision-dis...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cloud-vision-discuss/4ad305d5-14c8-42b2-adfd-98fe3a1a0657n%40googlegroups.com.

[David Yao]

Hi Jana,

advanced_ocr_options is a new field inside text_detection_params inside image_context:

[Osvaldo Salazar CS]

Hello, we are also interested in extending the term of the legacy model. Just today our system stopped working and after being online for over a year.

Following the alternative of using legacy_layout I have been able to make it work for now

Thanks

[Tianli Yu]

If you have tried "legacy_layout" in advanced_ocr_options and it works for you, you don't need to apply for an extension. This option is implemented in the new engine and will be continuously supported. This is our recommended solution.

The extension of using the legacy model is only needed when your pipeline absolutely has to revert back to the EXACT same result as the previous OCR engine. 

[Aplicaciones Coorporativas CS]

Oh thanks. I had understood that it was a temporary solution, yes, it did solve my problems and I will use that fix. If I have problems later I will let you know

Best regards

To view this discussion on the web visit https://groups.google.com/d/msgid/cloud-vision-discuss/edd24282-78e8-4d6b-b56a-e85501466c54n%40googlegroups.com.