Storing Auto Tagger keywords somewhere else than keywords field

[plastikman]

I have run Auto Tagger on all my images using Google Vision. Now I have one or more tags with Google Vision|tagname for each image. However, these can be numerous and "pollute" my manual keywords. Many of them are also wrong (which is OK and expected with maturity of this technology). But I still find it useful to have them for quick searching on terms I might not have tagged or missed.

Ideally, I would run Auto Tagger and store the keywords in a separate field where they don't interfere with my manual keywords. Photo Supreme has a separate AI category for this functionality, and my favorite Lightroom plugin AnyVision has a plug-in specific field in use for this, with an optional manual writeback to keywords (which also allows you to reassign keywords with new criteria from this database/field, without having to re-run in the cloud costing money). From the implementations I tried, the AnyVision plug-in by John Ellis has the best overall implementation of machine vision AI in a DAM package mainly due to configurability but it only works with Google Cloud it doesn't have access to the fantastic Thesaurus matching features from IMatch 2020 (and of course it needs Lightroom, which is a no-go for me!).

TLDR; I would like option to store machine vision AI tags in a separate field from keywords, so I can keep manual and auto keywords apart while still being able to search/utilize both within IMatch. Showing both manual and AI keywords fields on top of each other would also allow for easy transfer between manual (high accuracy) and auto (low accuracy) keywords.

A lower-priority nice to have for my workflow, curious if others think a like.

[Mario]

You want to store keywords outside of keywords? Not sure what that purpose would be.
IMatch could not use them and none of the keyword-related features would access these keywords.

You can use a prefix for all AutoTagger keywords so you can easily filter for them, remove them in batch if needed, etc.
You should employ the advanced features in AutoTagger to let only proper keywords into your database.
It makes absolutely no sense to allow useless keywords to pass through.

Or you could move all your unwanted keywords into a metadata tag you like using a Metadata Template or the Metadata Mechanic.

Keep in mind that most people want simple things. Like letting AutoTagger automatically add keywords.
People hate complexity. And storing keywords in an arbitrary tag is much more complex than most people will ever need. And custom write-back just for keywords. And extra effort to move keywords out of proprietary tags into proper keywords...I'm also no fan of proprietary XMP tags or custom namespaces. We have too many of these already and they always create trouble when you use multiple applications concurrently or you switch horses.

From users participating in telemetry I know that the average number of keywords per file between 3 and 5. Which is usually sufficient, unless you are a stock photographer or you process images for libraries and archives. But these don't use AutoTagger but carefully curated controlled vocabularies.

And if you already have two other applications which do what you want (PS and your favorite Lightroom plug-in AnyVision), why don't you just use these?
If these applications store the keywords in XMP subject or hierarchical XMP keywords IMatch can use them like any other keyword. This gives you exactly the functionality you want I think?
Specialist tools exist for a purpose.

[plastikman]

Thanks for the elaborate reply.

To illustrate my point: in the cloud version of Adobe Lightroom, I can search using Adobe Sensei AI for tags similar to the integrations that you have. However, it often tags my dog as a cat etc. But it is still useful to type "clock tower" and find images that I haven't tagged as such with so much detail. But because this type of AI makes many mistakes, I don't want all the mistakes show up in my XMP keywords.

It also has manual keywords which I use very simplistic and which are the ultimate source of truth since I tagged them. But with these manual tags, I wouldn't find the "clock tower" pictures because I don't manually tag that level of detail.

See how both can be useful? Of course going image-by-image and confirming the AI tags and converting them to Thesaurus manual tags would be the best option, but simply not feasible with 10k images in the short-term. So I still would like AI search available when I need it.

For now I threw out all my AI tags until I come up with another solution.  Again, it is not a priority for me but it would be useful. I just now purchased the software as it is, just trying to provide feedback on what I think would be useful.

[Mario]

I just now purchased the software as it is, just trying to provide feedback on what I think would be useful.

Which is very good and welcome. Keep going. I'm always listening.

But I also have to keep the big picture in mind. And "low complexity" is not something people would attribute to IMatch.
I hear "It's such a huge software with so many features. And so many options" way more often.

Throwing in "specialist" features like custom keyword handling outside the already 3 standard keyword storage schemes, with custom workflows and stuffs may help 5 users but confuse 500.
Don't get me wrong. IMatch is about power and flexibility and supporting different workflows. There are enough simple image catalogers or half-assed 'DAMs' on the market.

But this is one of the features which can become real ugly quickly.
For example if users suddenty want the "not keywords" produced by this to participate in things like versioning and propagation. Or wonder why these "keywords" don't show under @Keywords. And so on. Shiver.

During Beta we considered if storing the AI keywords into categories instead/in addition to keywords should be an option.
Because quite a number of IMatch users don't use keywords at all, but categories.

But then came the complexity into which category to put a keyword like "dog" - because many of these users have complex category schemata with multiple roots, identical categories on different levels, alias categories and suchlike. And this would require a fairly complex UI in the AutoTagger to basically map any incoming keyword to one or more category paths.
All in addition to the many ways you can manipulate and map keywords in the AutoTagger to integrate them into your existing keyword hierarchy and controlled vocabulary.

Doable? Yes.
Useful? For a handful or a dozen of users, maybe? No idea how many.
So I decided to not support storing AutoTagger results into categories for the 1.0 version.
And to wait if and how many users really request it.

If you ask users if something would useful, they usually say "yes". Even if they never going to use it. Just in case they might need it some day.
While it complicates workflow and option sets for all other users all the time.
So you ask them two or three times if they really need that now. And then a "must have" often becomes a "nice to have" or a "meh".

That's why this feature request board exist. So people can voice their ideas and others can chime in.
Even though the majority of IMatch users is silent and does not participate in this community, this board gives me a good impression of which features could be useful for more than a handful of others.

[plastikman]

Affirmative.

I am studying UX Design so I know all the pros/cons and considerations that have to go in feature requests and making sure they are balanced for different types of users. I also like to try out many software so hence all the examples from other software I have tried/use. But the fact that I bought IMatch after evaluating 5+ other options should tell you enough about my appreciation for your work.

I will give it some more thought over the coming weeks/months and maybe I come up with some idea and I will write it down in a more complete manner.