Please optimize the similarity search

[Mike]

For me at least that will be more and more important relatively soon. I will have to search large stocks regularly.

When I'm ready, I'll put together an extensive set of systematic image variations that will test the similarity search more closely so that I know where the limits are.

But even after short, unprepared tests with small groups of 300 to 1000 images, I had the impression, for example, that images that are rotated by 90 ° perform particularly poorly. iMatch is more likely to find completely different images than to recognize the similarity in the original rotated 90 °. The preferences didn't seem to change much about this problem. 180 ° rotations are not very good either, but not as bad as 90 °.

Is optimization already planned? Can we perhaps help to drive this forward in a targeted manner? E.g. by sending pictures that are badly recognized? Or do you have an other idea?

[Mario]

An image rotated by 90° will not be found by the similarity search. This search was build for the purpose to find similar images. Not looking for images rotated by 90° or -90% or flipped or mirrored.
It is also one of the least used features in IMatch. It is available since IMatch 3, and it does well for what it has been designed. It has a rather low priority for me. Since I have only one spare time developer resource, I concentrate on things actually used and helpful for the IMatch user base.

If you are in need of a better visual similarity search and you have exhausted all the options available for this search feature, I recommend licensing a dedicated software for this purpose.
When I recall correctly, Google offers something in this areas, as well as Microsoft, Amazon and other cloud providers.
I neither have the technical expertise nor the thousands (!) of developer resources these IT companies have.

This is not something I can "improve easily". This is an area of ongoing high-end research carried out by universities, start-ups and the big IT companies because they can monetize the results. Maybe you can find something in the open source. Else you will have to find a suitable subscription and service for your needs.

From your posts, I understand that you have a very peculiar and probably unique workflow. Maybe IMatch cannot do everything you want, or not always in the way you want it. That's normal. You may need to combine multiple applications and services to implement every phase of your workflow. From Windows Explorer to specialized AI image searches.

[Mike]

Thank you for the explanations and recommendations. I can understand the situation very well. I will look for additional options. No software can cover all areas. I have to rely on several anyway, but I try to keep the number of tools as manageable as possible.

As you assumed, I actually have a somewhat unusual and complex workflow. Some of it is exactly what a designer and photographer needs, other things are clearly beyond the scope, that's the scientific part. In the first area iMatch helps me perfectly, in the second it still offers me useful options, e.g. It complements my work with tools like Atlas.ti, a software for qualitative content analysis and theory building, etc., etc., very efficiently.

The only thing I don't like so much about iMatch is that I've only been using it since December ...

So it seems like a good choice