Note that part of the job could be done outside lightroom.
more and more camera have this feature, to adjust focus on the faces, even dslr (in live view mode). some of them store this info in metadata. So if lightroom could read that, this could help save computing time?
"From my experience : picasa is doing a much better job you Should try it. "
Okay, I tried Picasa on the same images.
It did much, much worse than Windows Live Photo Gallery, it was slower, harder to use, and very buggy. It missed a lot more really obvious faces it had no reason to miss plus it missed all the one WLPG missed too.
Took half an hour to scan the faces versus ten minutes, and would have taken at least 10 hours to manually enter them all versus 2.5 hours. I gave up and uninstalled it.
Interesting. I just gave Elements 9 a try. It did slightly better than Windows Live Photo Gallery, missing fewer of the ones wearing sunglasses. It still missed a lot of the profile and partial face shots, but it also ran by-far the fastest at about 1.5 minutes for the same set of images, and it only took about half an hour to manually add the ones it missed. Interesting, especially given how badly Elements 8 did on a completely different set of images.
My PSE9 catalog contains over 99,000 images. A large number of these photos are of youth soccer players from a variety of teams, and many photos contain multiple faces (e.g. team photos). As many players are unknown to me, it is very helpful to tag a face with a player's name, and to find all photos (e.g. action photos) containing that player.
Yes, at first PSE9's face recognition appears to work substantially better than PSE8. However, my experience is that it simply stops working after some number of images are processed. If you re-run Find->Find People For Tagging, nothing happens. Worse, other features are impacted: you cannot use F11 for full screen (must exit and restart app), and you must re-load thumbnails for many images (right click->Update Thumbnail, may or may not work).
It would be an excellent tool. At this point I use Picasa to detect and tag people but unfortunately the work around is not good for DNGs. I work events, set photography for films and publicity and face detection would save me lots of time. I even considered to migrate to Apple Aperture but the face detection in Picasa was better and the tools easier to use.
I have not used facial recognition tools and found the above conversation fascinating. My only reservations are that the technology does not appear to be reliable at this point and that LR still has basic stuff to get right, such as the keywording and performance rendering previews.
This seems to me to be an ideal function to be addressed by a third party addin.
- It is 2012, everybody has it, including PSE -> it is doable (probably quite easily as adobe has the technology
- It makes life easier -> that what LR is all about!
- It is not dangerous, as people already use keywords for it -> perhaps even more secure, e.g. a small checkbox "do not export personal metadata" on export
Please add this to LR. Even if it does not work properly it would be nice addition and start of a future development. This could be refined in each release like other features. Just the ablity to recognize a face and manually tag images would be great or search parameter "people in images".
Is it going to make it in LR4?
No idea but here is an interesting interview with Tom Hogarty on Dpreview
One feature that some users had hoped to see in Lightroom 4 is face recognition, in which the software identifies specific faces in images and embeds this information as metadata for easier image searches. Hogarty says that when allocating resources for a new release, highest priority is given to features that benefit the greatest number of users. 'Face recognition is very important to some', he says, 'but irrelevant to others, leading to [internal] debates about what solutions are tackled in a release cycle.' Perhaps even more important, he notes there are serious privacy concerns about, 'the ability of software solutions to collect person-specific information.' He says that the challenges in implementing a face recognition workflow in Lightroom involve: 'privacy controls, integration with third party solutions like Facebook, tolerance for false positives - and the effort required to correct them - as well as the time required [by the user] to teach recognition tools.'
So instead of having a full featured face recognition couldn't we have at least
-Region Tagging
-face detection (detect faces but not who does the face belong to)
This would at first greatly help the tagging process, with little recognition errors (less than trying to recognise who is who) and no privacy issue as the software does not automatically recognise people, just make it simpler for people to catalog photos with people on them (which photograph already do, today, anyway)
Regards
Eric
Well, interesting to hear this internal things. But also sad, as I really hoped to see that. As somebody else stated, the privacy issue is no issue to me: Now I do it manually, which is the same amount of information in the metadata but much much more work. Even a FR with say 90% accuracy would reduce my work by 90% (ok +- false positives and so on but hey, also 8% is nice).
I would even say that adding this function to LR will make privacy problem less important than today!
Why?
simply look at what they have done with geotagging
up to today I used to geotagg all of my photo outside lightroom, and when exporting them I had to option strip all of the metadata or keep all including private geotagged position. So my photo that were directly exported on the web had the position in them because I did not want to strip all of the metadata. (there are plugins that helps you do a better job - but not everybody knows about those)
Today LR4 let you define private position that will never get exported, or provide an option not to export any position.
Don't you think that the same thing will happen if face tagging is included?
We could add:
The thing is people always tagged (up to know using keyword only and not image region+keyword) people in photos and it won't stop. Auto tagging is just a tool.
It is like with knife
should we forbide knife because they can be used to kill
well no . This would be annoying for people that use them to cook, and people that may use them to kill will just use something else.
Should we avoid automatic face detection in LR
well I do not think so, this will keep "annoying" people that want to use it as a cataloging tool, this will not stop people that want to tag you on the web, as they will use other tools such as facebook, picasa, google+,...
regards
From Tom Hogarty's interview: "Lightroom involve: "...privacy controls, integration with third party solutions like Facebook..." Why on earth would we like to integrate our LR lib face detection data with FB. Why we can't just get a tagging tool with decent FR and leave it there. If someone decides to create FB plugin for integration that's an separate issue.
seriously ... WE NEED face detection in LR4 with a NICE user interface like Picaso 3 ....The interface in PS10 Catalog is terrible.... I have 30,000 pictures and on each one where it says "Who am I", I have to type the entire name....when the suggestions aren't correct (most time) ....whereas in Picaso, if I start to type Bob, when I type the letter B, it gives me all the names starting with B and I click the name.
The option should be an option to deal with people concerned with privacy issues....they can turn it off.
Facial recognition is NOT a consumer only feature but something professionals need also !!!!
Given that face recognition will probably not be in Lr4, I wonder if a 3rd-party plugin writer will step up. Its easy and fast to pull image data from the previews to analyze. If there's a command-line app to do the recognition, the plugin would be fairly easy to write... - The quality of results would be at the mercy of the command-line app of course...
Begs two questions:
- is there a good enough command-line app?
- would people use the plugin, or just continue to do without until native?
I took a brief look at the implementation issues a while ago. I think a useful plugin needs the following components:
- A user interface that lets users view and modify the rectangular regions of where faces have been detected, view and modify the people tags associated with those regions, and confirm the faces that are automatically associated with tags. Windows Live Photo Gallery and Picasa are two examples of such non-trivial interfaces.
- Face detection: A module that detects the rectangular regions containing likely faces. Detecting faces reliably is well understood, and the OpenCV library has code that does it.
- Face recognition: A module that associates a face identified by face detection with the corresponding people tag. In a quick survey a while ago, I didn't find any well-established open-source library that does this well. The best libraries appear to be commercial or proprietary inside Google, Microsoft, and Apple. Anyone taking this on should be prepared to invest a lot in understanding the application of machine learning to get good results; as a counter-example, Photoshop Elements licensed a third-party face recognition library but botched its application (at least in PSE 8).
- Code for manipulating the XMP region metadata as defined by the Metadata Working Group (this is not supported by LR, or any other application at this point, as far as I know).
I'm also disappointent to here Tom's response. AFAIK most/all competitor applications do have a face/object recognition. This is not only something that a few people would profit from.
Yes, we can do all the keywording manually, but it is time consuming. It is much faster to add location metadata to photos (by multi-selection) than to keyword their content as usually within a series of photos we don't jump locations that often. But we certainly do have multiple and different contents within a series (e.g. different people). So functional support on this side (keywording, face recognition or similar) would help to save much much time - more time than e.g. using a map module.
So please, Lr-Team, take this feature request seriously and don't wait until you've found the 100% solution as until then many may have already switched to a different software. We really do want Lr to remain a great application also for large amount of photos.
In Bridge, face recognition would be very interesting too. All people which use the Creative Suite would benefit from this...
Adobe could easily create a solution where the software recognizes the person as pattern, but doesn't automatically assign any names to it. It would be the user itself which takes the last decision. And the software wouldn't collect person-specific information, it would just find persons based on already existing images and ask for further steps. The information it assigns could also be less "deeper" and permanent, but just keywords which can be easily deleted as usual.
In fact this could be the same as our current work, just much much quicker!
In Bridge, face recognition would be very interesting too. All people which use the Creative Suite would benefit from this...
Adobe could easily create a solution where the software does find the person as pattern based on existing images, but doesn't automatically assign any names to it. A dialog would ask the user for correction and he would take the last decisions.
In fact this could be the same as our current work, just much much quicker.
I have read some of these comments and it is not completely connecting with me on how this will help my workflow if it is added.. unless it is added correctly and preferably even goes beyond facial recognition but object recognition.. as it is not just portrait photographers like myself using it.. but for a variety of image types where facial recognition by itself would not be too useful and if implemented I would want to be able to shut it off completely.
I HIGHLY disagree with one suggest of 'adding it because everyone else is doing it.. even if it doesn't work very well'. I trust Adobe will deliver good products.. I depend on their products to work well and not mess up my tagging or mis-tagging based purely on facial recognition. The last thing I need is Facebook like "Hey is this Gene Simmons in your photo?" or you have 239,234 photos in your library that have not yet been processed by facial recognition... would you like to do this now. It honestly becomes meaningless and a pain.
If something like this is added I would like to select a photo.. say "identify this item in the photo" tell the program it is a person or an object with potentially different algorithms if necessary. Then IF I would like it to identify and tag items I can choose which ones I want it to process like this.
Simply.. facial / object recognition is still an emerging technology at all levels from the phone to the super computer.. so unless it works well.. I don't need or want it part of my work flow... especially if I know who and what my items are already.
„adding it because everyone else is doing it” – It is quite improbably, that there is a relevant number of user, which are only asking for face recognition because of that.
User requirements can be very different. If you are working with a manageable amount of images and your workflow normally doesn’t confront you with an overcharging diversity, then it is understandable that you don’t have the same need for automated steps.
I’m managing about 300.000 images and some other things more. They all are as well part of design processes, but also material for many research projects. I can clearly say: there is not enough automatism for me! The many reliable classes and structures are automatically created, the less need for me to spend my time with stupid classifications. And you’ll surely understand that classes and keywords are a must in such a large database.
The similarity search has the potential to reduce the manual work. Face recognition is just a subset of the general similarity search and for a lot of users like me a great thing. I would like even more than that, automated classes like: human, man, woman, diverse ages and could imagine even many, many other like photo, canvas, drawing and so on...
I just hope that Adobe introduces as many reliable automated categorization features as possible. Especially for BRIDGE, because it accompanies all people using the Creative Suite.
Nobody would need to use them if not interested. Or do you currently feel forced to permanently read the complete Metadata? I'm sure, you just do it if you like.
„adding it because everyone else is doing it” – It is quite improbable, that there is a relevant number of user, which are only asking for face recognition because of that.
User requirements can be very different. If you are working with a manageable amount of images and your workflow normally doesn’t confront you with an overcharging diversity, then it is understandable that you don’t have the same need for automated steps.
I’m managing about 300.000 images and some other things more. They all are as well part of design processes, but also material for many research projects. I can clearly say: there is not enough automatism for me! The many reliable classes and structures are automatically created, the less need for me to spend my time with stupid classifications. And you’ll surely understand that classes and keywords are a must in such a large database.
The similarity search has the potential to reduce the manual work. Face recognition is just a subset of the general similarity search and for a lot of users like me a great thing. I would like even more than that, automated classes like: human, man, woman, diverse ages and could imagine even many, many other like photo, canvas, drawing and so on...
I just hope that Adobe introduces as many reliable automated categorization features as possible. Especially for BRIDGE, because it accompanies all people using the Creative Suite.
Nobody would need to use them if not interested. Or do you currently feel forced to permanently read the complete Metadata? I'm sure, you just do it if you like.
121
Replies
AdChoices