just curiosity: AI image analysis ?

[Michael Miller]

Hi Bill,
Yes, that’s my concern.  Even on a medium sized library like mine with about 26,000 images this would take about 3 days to process at 10 seconds per image.  A normal shoot for me is about 1,000 images, which would be almost 3 hours to process.

Cheers,
Mike

On Mar 30, 2025, at 10:24 AM, William Allen <dk at ballen.fastmail.fm> wrote:


On my 5 year old M1 MacBook Air one image takes 10 to 20 seconds. I suspect one could get some juice from parallel processing, but a few thousand images will still take a lot of time.

Regards,
Bill


On 29 Mar 2025, at 17:00, Michael Miller wrote:

Hi Bill,
I’m curious how long it took to process a single image?  I’m wondering if this could scale.  The other issue is converting the granite model to something digiKam can use.

I’m looking into this now.  Thanks for the idea!

Cheers,
Mike


On Mar 29, 2025, at 4:40 PM, William Allen <dk at ballen.fastmail.fm> wrote:

I have used ollama with the granite3.2-vision model to do this locally with pretty good preliminary results. Obviously it’s not integrated with digikam, but I think with some prompt tweaking one could get it to produce some usable and searchable content about images without exposing them to the world.

Bill

On 29 Mar 2025, at 14:26, Daniel Bauer wrote:

Hi,

I read now and then here about face recognition. For me personally this is not interesting, as my archive is ordered by persons.

But what would be really fantastic would be "general image content analysis" using *local* AI.

Like, for example, searching for images that contain
- a group of persons
- mountains in the background
- a person with a teddybaer
- somebody wearing a red shirt
- somebody sleeping/laughing/sitting/running
etc.

Of course, google, fakebook, X etc. can analyze the content of images this way, and very precise, but the (poor) tools available for the public only work with uploaded images, which, besides of all the data protection questions, means "giving them away".

A local AI (without uploading any data to anywhere) with such capabilities would be simply mind-blowing.

I have no idea if something like this is even possible, of if it is already planned for the future of digikam, but maybe my post here brings somebody to the idea, and who knows?

Have a nice weekend.

Daniel




--
Daniel Bauer photographer Basel Málaga
Twitter: @Marsfotografo (often explicit nudes)
https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.patreon.com%2Fdanielbauer&data=05%7C02%7C%7Cc7b9f30f40e043e1cacc08dd6f01f7ee%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C638788776433550753%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=VRb2TIPF8u6FLUCz0F15hzsDzPWRe%2BlF3kXLgHnpsPY%3D&reserved=0<https://www.patreon.com/danielbauer>
https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.daniel-bauer.com%2F&data=05%7C02%7C%7Cc7b9f30f40e043e1cacc08dd6f01f7ee%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C638788776433588997%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=JSbgG5Fh6b18S9hdTzVajrL3WrRNS39RfZMAD0fTyDw%3D&reserved=0<https://www.daniel-bauer.com/> (nudes)


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/digikam-users/attachments/20250330/eab93a37/attachment-0001.htm>