AI based keywording and asset managment

[kirk]

I finally tried new Lightroom  and kind of  impressed.   Not exactly with how its AI works  , it can't recognise my work related things well enough ( different kinds of textures and materials), but rather the direction in general.   It's what I consider  a trend of making soft truly  convenient . The one that would make you free from redundant puzzles, pro-active in offering you solutions and intuitive  instead of showing you nothing till you crack your brain figuring out  where you forget to check in  a checkbox or something.

I spent half of my life time studying what other people (software engineers) do rather than doing something creative myself.  Looks like soon it needs to be a special discipline : softwarelogy  like archeology  studying cultural layers of code in something like Zbrush or 3d max.
   Ironically in neverending race for  more features a pro level software  finally turns into an indecipherable mess.    I can't stress enough how  terribly, monstrously  inconvenient  most  CI art related soft are.  Well, it grants you your high salary because you know things  but gosh it's so huge pain  in ... usually.

With that new Lightroom I see a kind of new trend to make your pc an actual intelligent assistant and same simple to use as a brush and palette.  So I wonder if we have any chance to  see AI like things in iMatch?   Without cloud thing preferably?
         


       

[Mario]

I think this works only in the cloud-connected Lr version, not in Lr Classic, right?

IMatch already uses some AI features, but not in the general audience version for now.

I recommend my knowledge-base article Computer Vision in IMatch – Initial Results and the related threads about machine learning and AI here in the community.

Adobe more than doubled their revenue since enforcing the subscription model for many products. The subscription pays for itself, and the more cloud-only technology they add and make users depend upon, the better for them. If a user cancels the subscription, nothing cloud-based will work for him/her anymore. A superb lock-in mechanism. Kudos, Adobe!

Adobe spends a lot of work and millions of dollar each year in usability, to reduce support times. They make billions and have very deep pockets to pay for such things.

Every question a user asks in one of the official Adobe boards, via email or via one of the paid premium channels costs Adobe real money - because if the initial boilerplate auto-response AI does not help the user, an actual support person has to look into the case. Many questions are entry level questions and could be answered quickly by the user himself if he could be bothered to use the help system Adobe provides or one of the other resources. But many users (and at Adobe-scope this means millions of contacts per year) just ask their questions directly.

If Adobe can somehow make a feature simpler to use or setup an AI that takes the user by his hand, this will cut down on support cost. This can pay in full for setting up and training the Ai.

The base for these AI features is the "telemetry" data Adobe collects from all Adobe application while you use them. And that's a lot of data. Unless you have explicitly disabled all related features, Adobe can feed their AI with everything you do in their software., how you use the software, which features you use in which ways etc. This massive data collection is a good base for modeling an AI that supports the user. It would not be possible otherwise. All AIs based on neural networks need massive amounts of test data - which is why Google, Facebook, Adobe, Microsoft, Amazon Apple etc. all want your usage data and record everything you do. This data is pure gold (and free!) for business models based on AI technology.

Don't expect me to setup and train an IMatch AI to help you with DAM stuff in the foreseeable future. This is multi-million dollar technology.
I'm all ears when it comes to reduce complexity and make IMatch simpler to use. But without "dumbing it down", because there are too many entry-level "easy" DAM solutions / image catalogers around already. I also try to provide a good help system and training videos - which are not used by many users, unfortunately.

When you look at the questions asked here (or the ones I get via email) you can tell that many don't have simpler answers and I'm not sure that AIs will be able to help at that level...

[kirk]

Thanks Mario for your explanation,   iMatch is still best DAM out there.  At least I have not found anything better.   I totally understand the challenge

[Mario]

I sometimes have the feeling that Adobe is working on removing humans from the image generating and editing process entirely.
In a few years AIs may create better photos out of thin air than most photographers could ever achieve. And AI-editing and LR plug-ins will make them all look like Instagram or like another one these over-developed artificial HDR images.

If you visit photo competitions or exhibitions (I do) you may notice the trend that you see the same images all the time. At least that was my impression over the past year. Sometimes people even photograph the exhibits and them re-create them for the next photo exhibition, with slight variations. Boring.
And you can often tell which Lr plug-ins or process somebody has used to improve the image... (Still remember the Draganizer effect?) Or one of the many popular "single-click" Lr presets for family, wedding or landscape photos? I literally have seen hundreds of super-sharp focus stacked pictures of bees, ants, flowers and even sushi this year. But there is only so much originality in this. The editing quality always was good (thanks to RAW developer, Lr, Ps etc.) but I care a lot more for the motive. Maybe it's just me.

I'm pretty bored by all the hyper-realistic HDR photos composed in Lr, for example. Or the 'focus stacking' macro shots created by dozens or even hundreds of individual images. Boring. Of course I understand that there are very good reasons for HDR or focus stacking, especially when you are in product photography or architecture or landscape. But I'm really surprised these days when I see some new or original photography. And not again something that has been developed to death in Lr. And I look at a lot of photos each week...

I'll take a crappy, noisy but original or even unique photo all the time. Something that took some work. Not just an AI and a "one-click" finish.

Anyway, for a DAM, the nature of the images is not important. It just has to make sure that your metadata is clean and useful and that your images are organized well.
And that your image collection belongs to you, and is not locked into a vendor cloud somewhere.

[Jingo]

I'm pretty bored by all the hyper-realistic HDR photos composed in Lr, for example.

Unfortunately, this was bound to happen as high quality photography equipment was made available to EVERYONE... and not just "true" photographers.  Once EVERYONE had access, it only was a matter of time that software enabled images to be "fixed" with a single click.. and then Instagram came along...

[Carlo Didier]

Personally, I don't like all that AI stuff.
I tried the services that automatically generate keywords for images for example. That may get somewhere in the future, but right now it's like the very first software based translators. You had more work correcting the errors than you saved by letting the software translate your text ...
Or the first OCR systems. The corrections took more time than a good typist needed to just type the text in.

The worst AI like algorithm around is the one that Facebook uses to decide what might be most interesting for you to see. I always immediately switch to the "most recent" list view, because FB definitely has absolutely no clue whatsoever what I'm interested in 

[kirk]

Well. currently their AI keywording doesn't work for me either. It's just not intelligent enough yet.  But I love the idea of something would searching things  for me  and free my mind for my actual work.   All this without looking through my huge Evernote base  of "hidden" software gems and easter eggs I had to maintain for my countless computer graphics software tools as a reminder  of all those  "check this in" and  "find that in this roll down list" etc     

Somehow Adobe programs   need least space in that base so I guess they do something right. 


As of AI image editing   I would love to have it working for me too but unfortunately it's all seems focusing on hobbyist photographers and my job is not photography at all,  rather creating images for huge virtual worlds  through hand paint prototyping, sculpting, texture painting. nuance tweaking  etc.  Still lots of excruciating routine I would happily let AI to do if it can.    Sadly Adobe moves away from such kind of tasks rather for probably more paying back pull of hobbyist photographers and thus had to make their soft simple and convenient.       

For a few years already my main image editor is Allegorithmic Substance Designer. It's super flexible and potent but   gosh what a pain in the... it is. Like some alien race with a twisted inhuman mind made it.  But it's nothing new. All the pro level software is always that way.   

I sometimes think about paint tubes.  Such a simple and super convenient piece of technology some guy invented in 19 century  took many centuries before of  artists keeping their paints in pigs bladders for the very same purpose.    So hopefully in a few centuries CI soft might become same convenient too

[Mario]

As I wrote in several post and demonstrated in the state-of-the art KB article on my web site, the AIs often get it wrong. Sometimes really wrong.

Even if they would be 99% correct (not even close at this time) you would still have 1 out of a 100 images with wrong keywords.
Or 2,3....n with at least some wrong or missing keywords.

if you would base a search on that, you would always have false positive and, even worse, miss some images. Not good.

The AIs may be a big help for customers who face the task to add keywords to 50,000 or even 200,000 files. At least the AI keywords would roughly organize the images into some more manageable groups. And that's already a lot of help. For this specific purpose.

If your AI gets it only right in 95% of all cases, you will still have to check every image for missing or false keywords.
And in that case you are probably faster doing it yourself, and 100% correct. IMatch has many features which make adding keywords very fast.

Besides, many pro photographers often shoot the same type of motive, with very little variation in keywords (consider portrait, landscape, press, ...) and they can manage really well with a couple of metadata templates and an occasional manual keyword entry (e.g. the name of the product, person, politician).

Travel photographers have slightly different demands, but no AI can reliably find keywords for the track you are hiking or the streets or buildings you are visiting during your travels. These keywords need to be added by hand manually.

It's a hype and it looks good in the magazine articles. And when users actually put it to a test and notice that its not that good or reliable after all, the press trek has already moved on to the next round of products.

I will continue looking into this technology because I see other uses for this as well in a DAM context. But telling users "Let the machine assign keywords to your files, then go through all the images again to remove false keywords and add missing keywords" is probably not that smart. And I'm not Adobe / Apple so I cannot discuss this away and just tell my users that "If it works that way, it has to be that way. You are holding it wrong"...