Is there 'learning curve' at 'Faces' recognition ?

Question

Level 1

100 points

Is there 'learning curve' at 'Faces' recognition ?

My Library has 200,000 images, 90% portrait,
I started to use Faces , now Named approx. 5500 images.
but Aperture 3.1 suggest to me many many unsimilar faces.
If I continue naming work,
Will it improve recognition dramatically at any point?

MacBookPro 15,8GB,500GB, Mac OS X (10.6.4), Mid 2010

Posted on Nov 3, 2010 7:06 AM

Reply

Answer 1

Best reply

Warwick Teale

Level 3

584 points

Dec 14, 2010 8:55 PM in response to JUN48

yes Aperture will start to refine the faces "recognition".. Some tips from my own experiences of late
• for the project click on the faces menu (cork board appears with some faces on it)
• select any FACE
• you get (a) currently known images for that face and (b) MAYbe some others it has detected thet might be the same person.
• click on "confirm images" at bottom of window pane.....
• you get "confirm image" for each image in bother areas (a) [and in (b) if it found any for you to add or remove for tht person)]

sure you can go ahead and click on each one.. but when you have 100's this is tedious

now here's a quick trick:

either:
1) +select (marquis) / make a selection+ of those that you want and *they go GREEN*. Then simply click on those in hat new selctipon that you dont want and *they go RED for not confirmed* as the person... the later is very usefull for tell images that really are not for that person.

or
2) similarly *hold the OPTION key* down tha does then make the selection as above. This selects then all +for REJECTION (in red)+.. then simply go an click on the ones that you want.

do soem or all at once....

press the "DONE" button at lower right corner of pane when finished or when you have done a few.

repeat the above until all done.

*TRICK is here to identify those images that DONT belong to that image*. Soon Aperture will refine the FACES recognition selections I have found.

I have found this is very very efficient fpr getting tonnes of images into a single perform using faces.

hth

w
hk

Reply

Answer 2

gtminch

Level 1

0 points

Dec 18, 2010 1:36 AM in response to JUN48

I've been wondering the same thing. Aperture has just gone through my library of about 50000 photos, and its suggestions seem pretty random. I've spent hours naming faces, concentrating on a few quite distinctive ones and hoping that it would cotton on and start grouping them together.
But so far I see no evidence that it's hit rate is improving. When it shows me 50 suggestions for a person, there will be 2 hits at best.

Reply

Answer 3

gtminch

Level 1

0 points

Dec 18, 2010 1:37 AM in response to gtminch

Oh, and I've been using "reject" just as much as "confirm".

Reply

Answer 4

Dec 22, 2010 9:09 PM in response to gtminch

for faces: just a tip ... would suggest you try and use CONFIRM that the *faces are NOT the particular person* you are trying to tag it with.

_Simply this is excluding those faces rather than rejecting them._

You will find that this simple workflow addition will "teach it" I imagine.

Summary:
• select those faces that *+DO NOT BELONG+* to the name
• option click/click again to turn selection to RED (from green of black)
• confirm faces (bottom R.H. corner) that these faces DO NOT BELONG to the face.

This narrows it down over time and after many faces AP faces will make a much more accurate selction of a face.

try it and post back 🙂

w
HK

Reply