0.25
0.5
0.75
1.25
1.5
1.75
2
Tom-vs-Pete Classifiers and Identity-Preserving Alignment for Face Verification
Published on Oct 09, 20125552 Views
We propose a method of face verification that takes advantage of a reference set of faces, disjoint by identity from the test faces, labeled with identity and face part locations. The reference set
Related categories
Chapter list
Tom-vs-Pete Classifiers and Identity-Preserving Alignment for Face Verification00:00
How can we tell people apart?00:13
We can tell people apart using attributes00:19
Limitations of attributes01:04
How can we tell these two people apart? (1)02:20
Orlando-vs-Lucy classifier02:28
How can we tell these two people apart? (2)02:42
Steve-vs-Brad classifier02:49
How can we tell these two people apart? (3)02:58
Tom-vs-Pete classifier03:03
Tom-vs-Pete classifiers generalize03:46
A library of Tom-vs-Pete classifiers05:20
How can we tell any two people apart?06:12
Tom-vs-Pete classifiers see only a small part of the face07:25
Face part detection08:44
Alignment by piecewise affine warp09:04
Identity-preserving alignment10:58
PAW discards identity information11:47
Generic parts preserve identity12:36
Effect of Identity-preserving alignment13:34
Reference dataset for face parts14:14
Estimating generic parts14:54
Verification system15:38
Evaluation: Labeled Faces in the Wild15:53
Results on LFW (1)16:22
Results on LFW (2)16:46
Results on LFW (3)17:05
Thank you.17:11