Please share your template markup as well, it will help us to investigate your issue exactly.
Hi,
Pls find the template file and recognization files.
Thank you,
Shashi
recognization_file.zip (5.5 KB)
template_file.zip (590 Bytes)
@shashiakk
We have opened the following new ticket(s) in our internal issue tracking system and will deliver their fixes according to the terms mentioned in Free Support Policies.
Issue ID(s): OMRCLOUD-214
You can obtain Paid Support Services if you need support on a priority basis, along with the direct access to our Paid Support management team.
Hi,
I have taken a paid subscription. Can you pls let know the ETAs?
And, I am not getting result and that bubbled up options are not recognized as student id from a composite grid
All the files have been attached in my prebviousc
Pls look into this
Thanks for the additional information. I am afraid the issue is still pending for the investigation. As soon as the issue investigation is completed then we will share an update/ETA with you accordingly.
Hi Tilal,
We pushed our application to production. And we are waiting desperately for this fix.
Can you please do the needful in prioritizing this issue?
Thanks,
Shashi
Hi Tilal,
Below are my consolidated concerns:
- Eventhough the user not bubbled up, Aspose is returning as it was bubbled up and sometimes, Aspose is returning as two options are bubbled up.
- And, We created a composite grid that user can bubble up his/her id. But it’s not recognized.
i am attaching all the supporting files template, form.Omr, OMR generated and scanned file and the complete json returned from Aspose in a separate file.
Please let me know if anything need to be done from my side.
Hoping an early resolution as we purchased service.
Supporting Files.zip (1.0 MB)
Please note product team schedule the issues’ investigation and resolution on first come first serve basis. We feel this is the fairest and most appropriate way to satisfy the needs of the majority of our customers.
However, if you have subscription to our paid support service , then you can escalate the issue using it, as paid support has different precedence queue.
Hello, @shashiakk
Thank you for your patience!
I have investigated attached files. Yes, I can confirm there is an issue with default recognition threshold (40). Because of bold text inside of bubbles, there are a lot of false positives.
Because of different approaches to marking, we provide a configurable option - recognitionThreshold.
With current scan quality, text boldness and marking profile recognition threshold of 50 will produce best results, providing 100% accuracy to provided .pdf file. We have an article on manual fine-tuning
We are grateful for provided insight. We will plan an easier threshold prediction solution.
result.7z (634,7 КБ)
Hello @nikita.korobeynikov
Below is my use case:
- I am doing a POC in my local machine. All the supporting files are placed in my local directory. When i am testing this from local machine, it’s working fine.
- But in my application, i am storing the supporting files in AWS S3.
When i try to get the results of the scanned OMR sheets, results are not as expected.
I am doing the same as i am doing on my local machine.
I am uploading a scanned copy of omr sheets file from my web app, and converting it to base64string
in my node js api. And passing it to aspose api along with supporting files in base64string format.
Can you please let me if i am missing something?
Application-Supporting Files.zip (4.7 MB)
Local Machine-Supporting files.zip (4.7 MB)
Hello, @shashiakk
I have investigated recognition with attached files. There appears to be a misalignment, even in local machine files.
To further debug this issue, can you please help us with next questions:
-
What solution have been used to generate this template?
Aspose.OMR library: which version?
or Aspose.OMR Cloud: when? -
Can you share code that been used for this generation? Full function or requests for Cloud. Particularly interesting is GlobalPageSettings configuration.
-
Can you please share results of this template generation?
Images and .omr files or raw response body
Thank you!
@nikita.korobeynikov , Thanks for your reply.
Below are the response from my side:
1. What solution have been used to generate this template?
Aspose.OMR library: which version?
or Aspose.OMR Cloud: when? Answer: I am using cloud. I generated the files lastweek.
2. Can you share code that been used for this generation? Full function or requests for Cloud. Particularly interesting is GlobalPageSettings configuration.
Answer: Attching the file Code_To_Generate_template file where it has request and omr template code and settings in the code.
3. Can you please share results of this template generation?
Images and .omr files or raw response body
Answer: Attached zip has the images and .omr file. And same zip has the file containing final results.
Local Machine-Supporting files.zip (4.7 MB)
Hello, @shashiakk
I have repeated “generate-fill-scan-recognize” process with provided template using OMR Cloud. (see attachment generation-recognition-test.zip (1,6 МБ)) and was unable to find an issue with the results.
So I have compared my version of generated template image and your scan
Comparison.jpg (161,9 КБ)
There appears to be a difference in heights and there are differences in .omr files as well.
This could only mean that provided template configuration\template layout\image are not matching. This is a reason for incorrect recognition results.
- Please confirm that omr layout(“form.omr”)\generated template(“??.png”)\scanned image(“Newfill3.pdf”) are come from the same batch.
.omr file contains precise coordinates and is bound to the specific image with which it was generated (they both are returned by PostGenerateTemplate).
Any changed in template or template layour file is affecting recognition.
@nikita.korobeynikov , Thank you for help.
Can you please confirm below?
- Does it require to have each .png file should have separate .omr file?
- And after scanning, the respective scanned file and respective .omr file should be sent to Aspose cloud apis?
Thank you
- Yes. .omr file store coordinates for recognizable elements (grids\barcodes\bubbles).
To avoid complication, I would recommend matching each generated template with a respective .omr file.
We have a batch processing feature in the works at the moment to simplify this process for multiple similar templates(e.g. class of students).
- Yes, scans and .omr file should be used in strong matching.
In global setting i mentioned papersize as Tabloid. But i am printing it on A4 and scanning as A4 Size.
Will this configuration impact?
Find the attached folder. You can find .omr, template, .png and results. Facing an issue where some options are recognized where that is not bubbled.
Let me know if anything is missing still. Your help is much appreciated.
Thanks,
New-Recognization.zip (1012.1 KB)
Shashi
Hello, @shashiakk
I have investigated attached files. I am sorry to say that the reason for poor recognition was a bug with a “recognitionThreshold” parameter.
But thanks to your feedback I have been able to fix and republish a new version of OMR Cloud with a fix.
I have attached a PostRecognize request body with a recognition threshold set to 45. It will produce now a 100% accurate result, based on font boldness, bubble size, markings pattern and recognition quality.
PostRecognize-body.7z (653,8 КБ)
Thank you!
Hi @nikita.korobeynikov ,
Thanks for your support.
I am attaching below folder where it has 3 different scanned files of same original(of a same hard copy.). I am attaching the 3 different results also named as Correct, Wrong1 and Wrong2.
Can you please investigate the issue why it has different results? In correct-results.txt file, it produced correct results but in 2 other files, it did not.
Attached .Omr file which is same for 3 files. And attached template generation code also for you information. In the template generation code, you can replace the parameters as you want. May be you can hardcode the values in the template code.
Evaluation issue.zip (397.4 KB)
Thanks,
Shashi
Hello, @shashiakk
I have investigated attached files. Reason for incorrect recognition result - is different rotation of bottom and top halves of the page
See attachment below. Left is “Wrong-2.pdf”, right is “Correct.pdf”.
wrong_2-correct.jpg (227,6 КБ)
Top half of the “Wrong-2.pdf” scan rotated disproportionate to bottom half. “Correct.pdf” does not have this issue and produce perfect results.
In most cases it means malfunction in scanning equipment(e.g. ADF). Usually it is not an issue with the documents, but in Optical Mark recognition forms it will produce false positive results.
I can advise increasing the size of the recognizable elements (bubbles) to try to mitigate these fluctuations.
It is an interesting case. I will add investigation for this issue in development queue.