LinkedIn
Twitter
Hello app creators. Weโre working on a new optical character recognition (OCR) feature, and I would love to hear from you all about how you would use this feature in your apps. You can reply to this thread or shoot me an email at tony@appsheet.com.
Yes, this would excellent for our workflow, currently we scan a lot of forms containing Code 39, QR, and a text box. We have to press the scan button multiple times per form. It would be great if we could scan codes and text all at once and use something like regular expressions to match them to the right field. If you need help testing or developing let me know! 
@tony First: I got it to work 
Thank you for that useful tool.
Here are my impressions:
- If you get the Error: โCould not train your OCR model, please contact support@appsheet.comโ, the example image(s) is/are not good. Start with just one and add one by one.
- The example Images have to have a good background. An image like this wonโt work.
It also wonโt work if they have to less background like this one:
- After generating an OCR Model: When you add a column to your table and then hit regenerate, it will crash the App. Error: Object reference not set to an instance of an object. In the Columns Tab, there will be no columns. It also doesnโt help to delete the table and to add it again. You have to generate a new App. Workaround: Before changing the column structure, you have to delete the OCR model.
- LongText is not working. If you have 2 lines like in my example, you have to separate them in 2 columns. After that you can do a CONCATENATE() in another column to combine those 2 columns.
- In my example, there is text field and in the same line also a Date. Sometimes, it separates them correctly in 2 columns. Sometimes it puts all in 1 column like FL19/026209 11.07.2019. Seems to depend on the image captured.
- Date has to be Text type. It wonโt work with Date type. Error: This entry is invalid. But perhaps this has to do with our German Date format. Workaround: Use another VC to turn itthe Text Column into a Date type with the expression:
DATE(MID([DateColumnText],4,2)&"."&LEFT([DateColumnText],2)&"."&RIGHT([DateColumnText],4)) - Characters like โรผโ are sometimes recognized as โuโ.
- In the OCR Model I can see no images:
- You need to have internet connection to use the OCR. Am I right?
EDIT:
- Double Quotes are not handled correctly.
"SZ ""e-line"" MJwould be regognized as
"SZ""e-line MJ
@tony I can see images now, but no highlighted regions.
In my case there are numbers and decimals. However I found out, that it wonโt work well will decimal and number columns. I had to use text columns. Then the image was recognized better.
I am really interested in the progress of this feature, Iโve had multiple people reach out to me wondering if thereโs something in the works and I love to see it being tested.
Thanks to everyone for leaving your feedback and thoughts on your tests here, they really mean a lot and provide great insight!
I canโt wait for this feature to go main stream
@tony Now I can retrain again with the same data. Even with more data. Thank you!
But when I create a new Model, it will save the model, but without any result. It will not show any images in the patterns section. I think it hasnโt recognized anything.
I did this with just the same data i used before.
In one case it gives the warning:
A warning was produced while training OCR Model 'New OCR Model 2': The OCR Model builder found inconsistencies in the input data for column "CO". In row 0, the value "32" was expected but was not found; ignoring thisfield in this row for now. In row 2, the value "29" was expected but was not found; ignoring thisfield in this row for now. In row 3, the value "22" was expected but was not found; ignoring thisfield in this row for now.
But as you can see, in my data I have those values. But in other rows. Why is it expecting something in row 0?
appName=OCRWalk-133783
@tony I see you are working hard on this feature. Itโs getting smoother and smoother. Thank you!
Still having problems with LongText.
"SZ ""e-line"" MJ 12, Digitaldruck
PU-beschichtet
When I try using a LongText Column it catches only the first line.
So I split it in 2 columns. But then it rarely catches the second line.
Any idea?
appName=OCRKM-133783
Show Type Image with a GIF โhttps://media.giphy.com/media/sSgvbe1m3n93G/giphy.gifโ and with show_if condition AND(ISNOTBLANK([Image]) , ISBLANK([Any OCR column]))
To make it clear for the user that he has to wait a bit
