Facial Recognition work around for Capture One
I really like the refinements in Capture One 20 and they make it my favorite editor, but I need a way to take it up a notch to add facial and object recognition. I have QuMagie on QNAP (still testing it out), but it is cumbersome to have it in one app and then have to go to Capture One to try to get the photo pulled up.
Does anyone have a workflow that makes this smoother? Do you have a preferred AI photo analyzer that you use?
I feel like Capture One is pretty mature and powerful and it is now time for the developers to get the green light to focus on features to help us manage large collections of photos without having to rely solely on key wording. Imagine looking for a face and it finds relevant photos ACROSS all sessions. I quit trusting the catalog after getting burned and I wanted to avoid difficulty moving photos from the laptop to my desktop after a trip, so I became a sessions convert, but.... how do you manage across them as a collective?
I surely must not be the only one drowning in decades of photos... I have put off a massive indexing effort believing that some technology can help me some day. It is my only hope Obi-Wan Kenobi. ;-)
-
No comment til today? I am hoping for a workaround too. Indexing persons on every job is a massive effort.
No need to integrate, third party solution would help too.
0 -
I have been trying out the equivalent face recognition application on Synology having discovered it recently released.
I can see it might be useful for family and friends groups with a repetition of subject matter arising somewhat randomly.
However for other types of subject matter where a lot of faces of unknown people may appear it could be a annoyance. IMO.
At the moment I tend towards the opinion that it has more potential value to the cell phone casual shooter than to a more formal image shooting and processing activity since the structured "shoot" post processing allows one to select and name the subject with some precision at the time of processing/file naming/file output naming, etc. There is no need to have an application discover a whole load of faces and then ask you who they are.
It's also very much a "background" job - at least it is on the Synology box I have and I doubt that the nature of the activity, at this time, could be made much faster without somewhat greater and difficult to justify expense. That may, of course, change in the future.
The Synology application also attempts other categorisations. So, for example, it will identify Child, Infant, Sky, Forest, Garden, Park, Playground, Animal, Plant, Flower and Cake from a random selection of family activity snap shots and a could of Moon photos (Sky category content.)
In general the categories are reasonably correctly categorised.
In anticipation that one may change systems and have multiple storage repositories over time I tend to name the files and the output from my RAW processing to make basic identification easier in the future without creating a complex exchange of data requirement between image files. The secondary benefit is that if sharing the file with others the name alone can provide them with some indication of what, when and where.
Since I also have that name in the "Gallery" application(s) of choice they can be allowed to do the things that they do but still provide me with a file name that allows me to quickly find the original file should I ever need to go back to it as a RAW file and reprocess in some way.
Now the thing about all of this, especially in terms of face recognition, is that the software does not attempt to suggest who the faces are - at least not so far in my voyage of discovery. Nor should it in my opinion.
So it one shoots photos at events - or in the street - one might end up with a lot of faces discovered but no name to associate with them - but also no way to tell the software that you are not interested in that face or, to put it another way, are ONLY interested in certain faces (or near variants of them).
So one always has to take the time to identify the faces discovered - which is how it should be but it is entirely possible that many people will end up hundreds or possibly thousands of "Unknowns" and I can't see a logical way to eliminate that possibility given the nature of the task.
If one can crop or modify an image to eliminate the unknown people in the content BEFORE exposing the image to the AI processing is may help. But that really only works of you are running the AI against the final output image. Logically, therefore, the results are likely to more efficient and effective against an image Gallery rather than something like a RAW processing based application and its Catalogue.
Making a "Live" connection between the two may be something for the future but currently sounds like a "Plug-in" sort of application that, by one means or another, probably needs to work on the basis of a common identifier field - like image name - and then would require some well structured and stable data embedded in the images to make sure that the connection was not accidentally lost by, for example, name changes at some future point.
It all starts to get a little too complex, to my way of thinking, compared to the likely level of use (and degree of need) outside the requirements of the image agencies like Getty,
If your work involves a lot of head shot portraits there should be more effective organisational approaches to tagging the images. If shooting somewhat random group photos - perhaps at en event of some sort - there might be some benefit to letting an AI program find faces after the event and allow one to name them but at the moment I'm not totally convinced (based on the software I have available) that the AI process would be any more efficient over all then I could achieve using C1 and therefore adding tags to the source rather than just the output.
An application that comes up with suggestions for names to go with faces might be a privacy step too far at the moment. And in any case would be limited to faces - not the same as people recognition that one might want to tag for.
Identification of "type" of content (see above) has some potential but the level of accuracy (or perhaps innaccuracy) needs to be accepted. Either that or one would have to dedicate time to curating the results, assuming the chosen application offers such a feature.
Right now I think its a "wait and see how things develop" sort of situation. Unless someone decides to offer to develop a plug in application that could take on the task an work with an external AI application or Applications.
0 -
Sorry for delay. And sorry, I cannot follow your argumentation even appreciating your effort to write down your side of the things.
I do not know, what sort of photographer you are but maybe face recognition meets not xyur needs. Your imagination seems not to be not very strong. ;) And, just to say it, I am definitivly not an iphone photographer.
As I do weddings, birthdays, furnerals and more there are in every event some (let's say about ten to fifteen) persons I have to tag. Just for example: Not every friend of the bride, but her mother, father, brother, sister, uncle; so for the bridegroom too. And there are several hundred photos to tag. Face recognition would help to avoid a lot working hours a week.
About what you call "annoying": Nobody is forced to use even a single feature an software offers. So nobody should be annoyed. ;)
0 -
Thank you for your observation about my imagination, Ernst.
What would you do if a set of images were analysed and identified a number of faces that were not of interest to you and proposed that you might want to give them names? You don't want to name them because they are not of interest but it will still identify the area of the photo as a "face" and keep suggesting that you might want to name them. In that particular case for that software, as far as I can tell, if you want to use it at all you use all of it. There do not seem to be any parameters at the moment to tell it otherwise.
All you can do is name them "Unknown". Or something similar.
Now the software is not recognising a particular face - just an area of the images that looks like a face. It's quite good at that - so long as it can see 2 'eyes' and a 'mouth'. Especially if they look like they might be human.
So you add some names - 1 for each person in the shot no matter where they are in the image because it is recognising anything that looks like a face not an individual's face. Each ace is presented as In individual image.Therefore an image with 10 named people would mean the image appears in the gallery 10 times.
Add another set of images of the same person or people and it does not suggest names - just that it has found a face. For reasons of privacy that is just how it should be. If you want a police style Face recognition system I think you may have to seek some other sources - or at least I would HOPE that you would have to seek some other sources.
Now that software may not be the sort of thing you are hoping for but I used the example simply because I have been testing it recently and it is fresh in my mind.
I imagine you are looking for rather different system and probably one that works per project rather than as a single database for everything you toss into it.
So yes, you are probably not a typical iPhone user but that probably reinforces my point about needing some care to choose a product wisely.
It is not something that I would expect to find in a product that is mainly used as a RAW file processor. Nor in a camera, though the camera technology may also these days offer the ability to spot "a face" or something that looks like one, in order to offer the user some automatic tracking and focusing assistance.
Just my thoughts of course.
0 -
Thanks, SFA, for your care about my workflow. :) But we rather loose the path.
The starting point was: NN635976380434502912UL asked for facial recognition. I added that I would like to know a plugin or something similiar too.
The rest is not so important. I could tell you that I worked with facial recognition for years as I used Lightroom. Helped me to shorten my way to needed results (and no problem with images I didn't want to tag).Just believe me: I konow what I did and I know (in every detail) what I would do if C1 would offer such a feature. To explain to you in detail what I did when I came to the point that ... and so on ... doesn't change the game. Agree? Therefore it is not necessary to explain to me why I do not need what would help me to shorten my workflow.
I accept that C1 does not offer this feature(s). Capture One is not for me but for us all. That is not the point to be discussed. But I declare: Facial recognitzion would help me to shorten my workflow. GPS-tagging (with manual tagging during processing) is another helpful feature and not supported by C1 I accept too. I can live with. ;)
If your are an official speaker of Phase One - I have understood. If not - there is no problem for me to see that there are different views. ;)
Just my thoughts of course.
Kind regards
Ernst
0 -
Ernst,
It is rare to find anyone from C1 engaging in these forums (which have always been considered to be User to User information exchanges). The exception would be Lily updating the status of the Feature Request posts.
I abandoned LightRoom years ago (Version 1.4) so I will take your word for it that its facial recognition features save you time. My experiences of other software have so far seemed somewhat incomplete and, beyond some use in low volumes of subjects (i.e. mostly the same small group of people as in a family history), I feel that the C1 Keyword functionality is at least as effective. Maybe more effective overall. Even if it appears to be more intensive when working with it.
Of course we humans will usually prefer what we are used to (if we like it and feel affinity with a process we have developed sing the tools available) and most of us are not too comfortable with changes to our basic and long time used methods.
The Plug-in functionality, as it exists today, is mostly intended for 3rd party developers to create their own applications that can be integrated in some way with C1. It might be worth identifying an application that seems to fulfil requirements and asking them if they, or one of their partners, would be interested in developing something that would integrate smoothly and efficiently with C1. Or at least get them to talk to C1 to discuss things on a commercial basis.
Presumably any stand alone products might be usable in some way via an 'export - assess and process - update' model of some sort.
However given that what I have been working with in recent times is a newly released product on a significant, widely used hardware platform with its own processing capability, I have to wonder what one has to pay for something with more comprehensive capabilities (presumably possible from what I have read) and a faster process time which, in my opinion, would be required if the functionality was to be integrated rather than run externally and then batch uploaded.
I see some application developers are making claims of greater than 90% accuracy. That would be rather good levels of performance but still would not, in my opinion, relieve users of the task of checking the images for accuracy of identification.
Therefore I am not entirely convinced that 90% accuracy, good as it might be in computer processing terms, is really good enough for the sort of task we might desire.
I have a use for an application that can assess images and identify certain patterns as a very similar requirement to face recognition but simpler in many ways and so far have not found much that attempts it and nothing that succeeds well enough to make it worth pursuing.
Even if I found something that would address the requirement there will be some categorisation needs that simply would not work for certain images. In face recognition terms something like a side view shot where only one eye is fully visible and computer recognition fails whereas human recognition will succeed. Maybe that is not so much of an issue shooting typically somewhat formal wedding or event shots of people but for other subject matter the rate of automatic identification might be less of a time saver.
I note that one of the application developers observes tat they can achieve better face detection and identification from video than from still images. Maybe we should all shoot high resolution video and just pick the frames that work best? ;)
Thank you for the discussion.
Grant
0 -
Hi Grant,
I see your retentions. For me it is clear in every case - with or without face recognition - that I have to tag manally in most cases. Face recognition would just help me to tag faster. The time face recognition needs to get results is not so important because it does not need my working time. And any accuracy from 80 percent up is welcomed.
Why I am not looking for a separate standalone app? Because there is a database within C1 holding all metadatas. This database is the target of the tags produced with face recognition too - wherever the tags were produced. Some minutes more time to get them there. This is the reason I would prefer a plugin solution.
I can use (and I use) face recognition for older cameras with Lightroom (and hand over per XML. Time consuming but works. The problem is that more and more recent cameras will deliver the images. And Lightroom is on this side at an end in my workflow.;)
We can stay at the point that some would appreciate face recognition and some will not. For me there is an additional point that I prefer to have features I do not need rather that not to have features I would like to have. And - this is for every feature - I will find my way to work with every feature that is able to shorten my way to perfact pictures I need to make money with. I'm never looking for reasons why it COULD make troubles to use a new feature. I'm always looking for the best way to integrate in my workflow.
Anyway: Was an interesting discussion. Thans you!
Kind regards
Ernst
0
Post is closed for comments.
Comments
7 comments