Resource For CV Descriptors Advanced Configuration

We do a lot of automation via CV and have had to figure out how to work with descriptors via trial and error. Is there a resource somewhere that has detailed information on the available descriptor components and their usage?

Example, “target” can be Text, AnyText, AnyWordGroup, Input, Icon, etc. None of this appears to be in the official documentation.

Also, at runtime there are descriptor items visible that don’t seem to be documented anywhere.

For example, is there a way to force CV to use a higher level of accuracy than the default?

Hi @jason.petrasich

Till now we’ve been trying to keep things as simple as we can for the RPA dev, handling most things in the background, without creating the documentation of types and everything else.

Would you be interested in having a call with us to better understand what you would need and see what we can do to help?

For example, is there a way to force CV to use a higher level of accuracy than the default?
Yes - for Image or OCR target / anchors there is the possibility to change the accuracy level.

Hi @jason.petrasich, yes, as @sbotan was saying, we wanted to keep things simple, but you’re making a valid point. We’ll work on adding documentation for power users as well.

Let us know if you’re interested in having a call - we’re also very eager to find out your pain-points and whatever other feature you’d be interested in for CV. You had us from “We do a lot of autiomation via CV” :heart_eyes:

Yes, please @Bogdan_Sultana . I PM’d Stefan but i haven’t heard back. If you need my contact info, please PM me.

@Bogdan_Sultana Hi! The subject is interesting for me as well. I have a case where I need to check if some text is visible in a list, but the “CV Element exists” creates a lot of false positive results.

Hi @krzysztof.szwed

Try adding Accuracy:0.7 in your descriptor:


You will then notice a new section appears if you click Edit on Screen, Advanced Options and then click on the target on anchor you’re using (“Accept” text as an anchor in my example here):

0.7 is the default value for Text, so you can try using a higher value to get rid of the false detections. (0.4 to 0.95 is the allowed range)

Let me know if this worked. Also, did you have similar issues with other UI elements in CV? Would you use this for other than text elements as well? I’m trying to get some feedback on how useful would an Accuracy slider be for all types of UI elements.

The Accuracy for text works like a charm. Is there a similar or other attribute to increase the fidelity of CV Extract Table?
At the moment it is often skipping lines, either at the top/bottom of the table and/or when scrolling down (most often) to capture more data.

Great, I’m glad it worked.

No, we only have this for Image and Text. Would you find a fuzzy threshold slider useful for all CV elements? Or is it something strictly related to tables, from your experience?

We’re gathering user feedback on this type of how useful a more advanced control would be, this is why I’m asking. On one hand, we want to offer a clean UX, in which AI “just works”, but on the other hand there are a lot of corner-cases where our AI model fails to detect or hallucinates, so a threshold slider for the confidence would help…? What do you think?

I would definitely say that is has more uses than just for text. The issue I’m something having with tables is that headers are not always recognized 100%, which makes working with them a pain. eg. I need to do things like
( MyTable.Columns.Contains(“ColumnName”) AndAlso CurrentRow(“ColumnName”).Value=“xyz”) OrElse (MyTable.Columns.Contains(“ColumnName”) AndAlso CurrentRow(“ColumMame”).Value=“xyz”)
A great feature would be to modify a table’s headers to match an expected set of headers using Levensthein or other algoritm (of course I can also do it myself)