Feature request: Automatic highlight reels via OpenCV?

[Paul A. Norman]

While not dismissing out of hand some possible role of AI assistance here, perhaps like avoiding irretrievably over/underexposed or utterly blurred photos. 
And while guardedly seeing some potential in what you suggest, and not wanting to dismiss it out of hand, there are I believe deeper directional changes and developments to be explored.

What you raise here Bryan is going to become an ongoing paradigm question for humans in many realms—which I hope we are going to be aware and wise to quickly.

Just because it appears “we can”, should we?

Beauty is in the eye of the beholder – or actually the brain and heart behind the eye. Can AI which itself does not actually 'think', through extrapolating LLMs through rule based criteria, behold and select what humans (or a select group of humans) perceive as beautiful, interesting, and pertinent etc.?

Where is the barrier between human choice, recognition abilities, and thinking verses artificial machine so-called “thinking”?

When are we just slipping into letting AI ersatz and do our own thinking for us? 

Would we notice the transition?

There are already programmes to assist those (mostly young people) who have become *so* glued to and reliant on asking for electronic answers on everything, that they can barely hold fluent conversations with humans, or put two thoughts coherently together.

Can we realise when we have crossed over into doing no actual thinking, nor making any distinct personal contribution to our AI assisted, or as time goes by —  “AI governed” ventures?

Would others start to even unconsciously suspect a vanilla or bland or even automated approach to the selections they believe we are artistically presenting to them?

‘and certainly wouldn't replace human-centric selection of the perfect pose and expression’,

– Paradox –

‘... knowing they capture the gist of a folder.’

It becomes a kind of blind 'trust' or 'faith' that AI does the same, or better, job that we would do ourselves – especially in areas of discernment, insight and qualitative appreciation.

Other than by blind trust, we wouldn't “know” that the AI selection captured the “gist”, if it has done ALL the selection; and to “save hours of manual labor” — we never even appreciatively look at largest thumbnails of each folders’ contents?

Where then is the “flair” that people have come to associate with your personal choices?
— Shall we call it the peculiarly personal human element? Even the things that come from your heart—touching upon the spiritual side of human existence, that “artificial” – can not approach?

In auto preparing, using AI and coming to avoid hours of manual labour —

“.. they could just be copied directly to a slideshow, photo frame, or email, knowing they capture the gist of a folder”

— Could culturally dodgy or dangerously offensive results just slip through “copied directly”?

And at what future point does actual human review dwindle into abayence?...
“Ideally, Highlight photos could be selected to be further manually tagged, starred, labelled, or otherwise entered into existing workflows.”

If AI does all of the preselection of what even makes it before human eyes to be “further manually tagged, starred, labelled, or otherwise entered into existing workflows” – what is left behind and never seen—and is effectively lost?

All of the use of AI for bulk processing and preselection assumes that it (though not able to actually “think” or “appreciate”) – it is doing the same job as a human artist would do in assessing available material.

So ought we at all, inadvertently or otherwise, develop a blind trust or “faith” in artificial powered ‘non-thinking’ — in any human creative or “life” area?

Given that some now actually worship and dangerously(!) literally receive serious life counsel at the feet of AI “god” interfaces(!)
—then even small steps of AI adoption and adaptations must be strenuously evaluated  to see where they inadvertently, or otherwise, take us. 

We must not passively drift.

I look wider than you have proposed — believing we must become very vigilant as to where “artificial” is taking us in our genuine human expression, experience and indeed very existance.

Paul
---
https://PaulANorman.info


On 1 January 2025 11:47:56 am NZDT, "bryan at gillson.net" <bryan at gillson.net> wrote:
>Has the team considered using OpenCV for more than facial/object
>recognition given the recent improvements to the AI pipelines in
>8.5/8.6 (thank you Mike!)?
>
>I'm hoping for something similar to iPhoto's auto-generated highlight
>reels where DigiKam would identify a quick-and-dirty "good enough"
>sampling and representation of the contents of a folder. This would
>save hours of manual labor: Just selecting vacation highlights can be
>incredibly time consuming and intimidating when the folder contains
>well over 1,000 photos from my wife, children, and I with many
>overlapping subjects, duplicates, errant or "failed experiment" photos,
>etc.
>
>These Highlights don't need to be some Platonic ideal "best" photos
>(whatever that means), and certainly wouldn't replace human-centric
>selection of the perfect pose and expression. However, qualitatively
>decent pictures could be selected using AI-based criteria such as
>focus/sharpness, visual clutter, uniqueness, and rule of thirds
>composition. I don't need 100 elephant pictures from the zoo; just pick
>a few good ones. Just pick one of the 6 group pictures (even if
>someone's eyes are closed, I can always refine it later). Skip the 15
>pictures taken of the inside of my pocket. Et cetera.
>
>Ideally, Highlight photos could be selected to be further manually
>tagged, starred, labelled, or otherwise entered into existing
>workflows. Or, they could just be copied directly to a slideshow, photo
>frame, or email, knowing they capture the gist of a folder.
>
>Would others find such a feature useful? Or is DigiKam already capable
>of something like this and I'm just not aware of it?
>
>Thanks!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/digikam-users/attachments/20250101/6d0f4c4d/attachment-0001.htm>