Software for AI picture indexing - help wanted!

Hey,

I have been taking digital pictures since 2002 and have approx. 500K of them archived in various external HDDs. Finally I have decided to index them into a searchable, usable gallery! Since it will take a lot of time to sort them manually, perhaps you could suggest some software or workflow solution.

Basically I want to:

1. Add face recognition and have them searchable by people featured in pictures
2. Add specific scene descriptions and keywords into Exif metadata (e.g. sunset; building; ship; car etc.)
3. Be able to search image EXIF metadata

I do not want to upload the pictures to any online service (takes time; storage will cost; privacy issues). I want to have a searchable indexed gallery on my home PC.

So my questions:
1. Does any software have face recognition capabilities?
2. Does any photo management software has AI for scene / object description and can automatically update EXIF metadata?
3. What is the best tool for Exif metadata search?

Or simply - how would you address my situation? What is your workflow for managing/tagging your gallery backlog?

I came to realize that there are so many moments captured in the pictures and due to the fact that the pictures are simply arranged by filename it is difficult to rediscover them.

Thank you so much for your support!

Kind regards,
John

Adobe Lightroom Classic or whatever they call it now now has face recognition software, but I have not tried using it. I pretty much have to do all of my keywording manually, since an AI is not going to be able to do it in enough detail for me. I don't think even the worlds best AI could tell the difference between a Hawker Hurricane Mk II and a Mk XII.

Lr CC has Sensei which is supposed to be a usable AI for sorting images by content. Again only going by reports from Adobe themselves.

Alan

I am looking into face recognition for a while and have not found a sattisfying solution so far. There is Lightroom, which comes closest to your needs. I used LR 6 for face recognition. Works decent - but is/was an extreme ressource hog and the recognition was fair at best. Microsoft has an onboard solution (MS Fotos?) for windows machines that also comes with face recognition. It worked well with a few pictures - but crashed faced by the daunting task of dealing with tenthousands of pictures. Not sure if this was a local issue on my machine only.

The by far best and most robust face and scene recognition AI is the one of google photos. I have a subset of my pictures up there and are absolutely amazed about the accuracy of their AI. It groups faces automatically without training with an astonishing accuracy. Really. Additionally, it does tag different scenes (holidays, hiking, ....) quite well. But as you stated: not for you as all the pictures need to be in the google cloud.

What's on my list is TagThatPhoto which is brand new and is a successor of Fotobounce. It's out for a few months and I did not have the time to test it. Alternatively, I am thinking of making my own classifier, using some open source AI deep learning algorithm. Takes time though....

(P.S.: If you can differentiate a Hawker Hurricane Mk II from a Mk XII on a picture then an AI can be easily trained to do this. Trust me - this stuff is nowadays scary reliable. We had a project where we trained an AI to recognize manholes from Google street view pictures with astonishing accuracy.)

macOS Photos app has served me well in regards to AI. I don't use it for professional photos, it's far too limited in terms of editing. But if I search for something, it generally turns up stellar results.

drmaxx wrote in post #18896827
(P.S.: If you can differentiate a Hawker Hurricane Mk II from a Mk XII on a picture then an AI can be easily trained to do this. Trust me - this stuff is nowadays scary reliable. We had a project where we trained an AI to recognize manholes from Google street view pictures with astonishing accuracy.)

I picked that example because the only difference is that the Mk II was built in the UK, and the Mk XII was built in Canada. So visually they are identical. To identify them you have to pretty much look up the identity of the aircraft online. Doesn't help that they repaint them in different schemes from time to time, and even disguise the versions for things like films every now and then. What makes life easier for the Hurricane aficionado is that there are only currently 9 of them, over all of the variants, airworthy in the UK. There is at least one in the USA but I think that's about it. My usual advantage when tagging images from an airshow is that I at least have a list of the types that took part, and the order they appeared in. I suppose it won't be that long before the AI can recognise the basic aircraft outlines, analyse the markings, and go look them up online for you. At least it should be good at scanning the various national registration databases, since that would only require simple translation skills.

Alan