How to use face detection

[tstoddard]

Until now, I've never wanted to use face detection. There's something scarey about computers recognizing me that I'm a little uneasy about. But, I see the potential of using it in a DAM to help speed up labeling, especially when there are a lot of photos of family members like on a holiday. Right now, I don't have any unlabeled photos of people so I don't have any fresh files to experiment with and I don't know where to look for help. I don't see any explanation of how to use it in the quick manuals that I reviewed but I haven't looked at all of them yet. Is there a reference that explains how it is supposed to work?

All I really want to be able to do is ask PSU to detect faces and automatically label the ones that it recognizes. I don't really care about areas, I just want the files labeled. Can I somehow have PSU look at faces in an image and then assign labels based on the faces in the image without creating "areas"? If anyone can give me some tips as to how to use this functionality, I'd appreciate it.

[DirkS]

All I really want to be able to do is ask PSU to detect faces and automatically label the ones that it recognizes.

Important thing to know is that PSU can do face detection, but not face recognition, so automatic labelling is not possible.

Gr.
Dirk.

[Hert]

Typically use Face Detection as follows (for single images):
1. Open image
2. Press F to detect faces
3. Drag the catalog label to the face area; this could be a label in the Catalog Explorer or even an already assigned catalog label in the LAP. The other way also works; dragging a face caption to a catalog label

[tstoddard]

Dirk, thanks for pointing out the distinction between face detection and face recognition. I guess I had hoped for face recognition. I thought of it as an efficiency booster but instead it is additional work. I don't really see the benefit in doing it for my purposes. Usually I'm only labeling people that I know, so I don't need to be reminded as to who they are, I only want to be able to find pictures that I've taken of them. I guess I can understand why some people might want to do this but I can't see a compelling reason for me to do so. Am I missing the real benefits of face detection? Why do people use it?

Hert, one question I still have is what happens when I press F or right click and choose face detection from the Area submenu and those "New Area" rectangles pop up all over the screen? Will those areas be saved even if I don't label them? If so, do I have to delete the ones I might not want? When I do it, I see an action panel open and close quickly as if the file is getting synchronized as soon as I do it. This leads me to believe that I have just written a bunch of new areas out to disk.

Thanks!

[Hert]

Tom,

Areas are part of the metadata standards and will be written in industry standard format. If you don't want areas then you should indeed delete them.

[freudenthaler]

PSU can do face detection, but not face recognition

Thanks for pointing that out! So what's the use case then for face detection? I can't see any benefit yet..

[snowman1]

Further to Tom's post, it would be great for back-converting images to be able to run detection on a catalog label (or folder or date or whatever) at a time; and have all the images in which faces are detected placed into a label so I can them go through them and assign the names to the areas. An obvious approach would be for the inital search-and-detect to allocate a dummy name to all the areas; I could then click on the dummy name to bring up all images assigned to a dummy name and assign them to the real name (or unassign them if the face detected was not someone I was interested in, just a passer-by say).

Is this possible? Do I feel a Mantis feature request coming on?

[Hert]

Face Detection is also available in the Batcher

[Hondor]

Usually I'm only labeling people that I know, so I don't need to be reminded as to who they are

There is the ever so slight probability, that your photos will exist longer than yourself. Consider proper tagging a gift for your descendants. My shelves are full of inherited photo albums with pictures of relatives nobody now remembers the names of. Apparently my ancestors felt the same way about needs to be reminded as you do. Most annoying!

[gcoupe]

There is the ever so slight probability, that your photos will exist longer than yourself. Consider proper tagging a gift for your descendants. My shelves are full of inherited photo albums with pictures of relatives nobody now remembers the names of.

Exactly.

[Andrey_Ra]

There is the ever so slight probability, that your photos will exist longer than yourself. Consider proper tagging a gift for your descendants. My shelves are full of inherited photo albums with pictures of relatives nobody now remembers the names of.

Exactly.

I found this argument strange. In comparison with the old times, now every certain person have from 100 to some thousand photos in our archives. (In case if this is important person in your life, of course.) And about 10% of these photos are the personal photos. Even if you are Mr. Pinkerton and have no thoughts about who is who - you might do the reverse identification without help of any kind of face areas.
Moreover - I think that this there is no hope that in future you will find these areas because users of PSU do not have any benefits from the face detection process. They just spend additional time without any advantages. Conversely, face recognition could save a lot of time in the process of labeling.
PS. I wonder why face recognition is not included in product even as licensed technology. You made lot of work and then stayed in one small step away from success

[Mike Buckley]

There is the ever so slight probability, that your photos will exist longer than yourself. Consider proper tagging a gift for your descendants. My shelves are full of inherited photo albums with pictures of relatives nobody now remembers the names of.

The chances that anyone in my family will learn how to use software to look up the name that applies to a particular person is slim to none, especially if there are several people in the image. If I wanted to ensure that people know the names of people in my photos, I would have to print them and label them on the rear of the photo. Example: Clockwise from top left: Jim, Ann, Paul, Frank, etc., etc.

[Hert]

Areas and area titles are stored in the standardized format as defined my the metadata work group. Most serious management tools today are already able to read this format; even a consumer tool like Picasa can write its face areas to the same structure. In a few years from now, about every tool out there and every sharing site out there will read areas. It will become just a common as keywording is today. Having your areas in metadata is very good practice, though I understand that some of you see it as unneeded extra work today...just like we all though it was for keywording a decade ago.

[gcoupe]

PS. I wonder why face recognition is not included in product even as licensed technology. You made lot of work and then stayed in one small step away from success

Because it's currently very expensive? Google and Microsoft can afford the investment required. Small software houses can't.

[Mike Buckley]

It will become just as common as keywording is today.

Considering the number of photos taken everyday, assigning keywords to them is very, very far from common practice.