Computer Machine Learning to Organize Photographs

[bovk]

You may find this interesting. If it does not belong here kick it out. The future of out photo storage:
LIFE Magazine Tags - Browse through the 20th century via Tags defined by Machine Learning : https://artsandculture.google.com/theme/kwKSLHCd3edAIg
and
https://artsexperiments.withgoogle.com/lifetags/

[Michael Erlewine]

Absolutely interesting IMO.

[bovk]

Thanks, Michael. Yes, I agree. In the past it was neural nets, rule based systems, pattern recognition and it was all kind of normal. Then came face recognition and voice recognition and it started to be complex as is Nikon 3D metering and AF. But the face recognition is still of recognition of one object and it took separate CPUs to implement. Now it looks like they take an object (like banjo) and identify it in all available photos, then take next one (like musician) and so on, resulting in a tagged database of photos.
Once I have an software like this I will stop using sub folders !
Bo

[Seapy]

The pages to which you linked made no sense whatever to me, not to condemn them for that!!! 

However, Apple have been doing this for some time with face recognition to a reasonable degree.  Lightroom CC (web based) relies entirely on AI to locate images by using freehand keywords, (not pre-defined) That seems pretty good but far from infallible.  I did dump a large number of my general image files into the Lightroom CC 'black hole' and to be honest I was surprised how retrievable they were, completely without allocated keywords, I imagine that will only get better.  When it can name obscure species or perhaps star groups... with certainty then my interest will be re-kindled!

As things stand, I know that if you were to dump all your images into either system, many images would never be seen again!

If you add some of the more 'artistic' offerings shown here and elsewhere, I cannot see how most of them could be retrieved, ever.

All my images are in folders by year, date yy-mm-dd and shoot  and will stay that way...

[bovk]

Robert - you mentioned many good points.
As far as the ref links; if you open the first one and scroll down and click on Discover the LIFE Tag Experiment link. After it is done going through the pictures click on START. Then I searched for "banjo" and clicked on a picture. That showed the tags for search.
I think this is a good visual example how it works, simple in concept, interesting when applied to large amount of photographs. 

Part of Nikon metering may work similar way: current image on sensor gets tagged, similar images are found by a tag search and the typical exposure is estimated by some method (probably not  a simple average)
I am keeping my sub folders for now
Bo

[Seapy]

I forgot to mention Lightroom Classic, (desktop) is now providing an AI 'auto' adjustment for images.  I was skeptical at first but I am using it a lot now.  I think that probably uses a similar method to the one you describe for Nikon, I think it has a lookup table of some sort.  My only quibble is the brightness seems to be too bright almost every time.  I wonder if it will learn that I turn it down for almost every image!  LOL  It is a great time saver though.  It doesn't yet do white balance or clarity, probably will in time.

[David H. Hartman]

Once I have an software like this I will stop using sub folders !
Bo

Run Silent, Run Deep?

or

We all live in a yellow submarine?

---

Software that could ease the pain of tagging and finding photos would be very welcome.

Dave

[arthurking83]

I've been using a program as simple and simplistic as Windows Photo Gallery since about the last 10 or so years with facial recognition.
In the beginning I dismissed it as cheap thrills, useless bloatware .. but after Nikon stopped support for ViewNX2, I had to look for alternatives to keyword/tag my NEF files.

Problem now is Microsoft have also stopped support for Photo Gallery as well tho .. so I'm back to where I started.

Reason for using Photo Gallery is that it embeds the keywords/tags into the NEF file(my preference), and once done, it's then synced with all my other software so that the NEF file's keywording/tags are seen via all programs used to open the NEF file.

For a long time I hadn't used the gimmicky facial recognition part of the Gallery software thinking it's silly, but about 3 years ago I had many old photo scans of my mother, and began to keyword them.
it then offered to use the auto facial recognition tools, located images of mum, and asked me if that was 'mum'. Accepted, and tagged those that were mum .. job done easily, and much quicker than I having to sift through images to find.

Tool made me laugh tho when I had an image of my daughter with he cabbage patch doll, and it located my daughter well enough but then asked me who the cabbage patch doll was.. gave me option for mum, my brother, and a host of other people I'd located earlier.

[MFloyd]

I forgot to mention Lightroom Classic, (desktop) is now providing an AI 'auto' adjustment for images.  I was skeptical at first but I am using it a lot now.  I think that probably uses a similar method to the one you describe for Nikon, I think it has a lookup table of some sort.  My only quibble is the brightness seems to be too bright almost every time.  I wonder if it will learn that I turn it down for almost every image!  LOL  It is a great time saver though.  It doesn't yet do white balance or clarity, probably will in time.

The new "Auto" is a major improvement. Overexposure was more the case with the former version; the only thing which doesn't work too much is "contrast" which is systematically lowered. And yes, "clarity" is not implemented (yet).

Another noteworthy thing is the AI (Adobe Sensei) used to find pictures in Lightroom CC (not the Classic one). You can find pictures by description without need of keywords. Quite impressive.

More about Adobe Sensei:

https://www.adobe.io/apis/cloudplatform/sensei/senseifunctions.html