0.25
0.5
0.75
1.25
1.5
1.75
2
Face detection without bells and whistles
Published on Oct 29, 201415722 Views
Face detection is a mature problem in computer vision. While diverse high performing face detectors have been proposed in the past, we present two surprising new top performance results. First, we sho
Related categories
Chapter list
Face detection without bells and whistles00:00
How to get a good face detector?00:07
Main points00:11
Benchmark issues00:27
Most relevant detection benchmarks00:32
Suspicious curves - 100:53
Suspicious curves - 201:08
Suspicious curves - 301:18
Top scoring false positives - 101:22
Top scoring false positives - 201:42
Issues with the current benchmarks - 101:50
Issues with the current benchmarks - 202:18
Issues with the current benchmarks - 303:04
Issues with the current benchmarks - 403:57
Solution: Improved annotations - 104:05
Solution: Improved annotations - 204:18
Solution: Handling different policies - 104:45
Solution: Handling different policies - 205:09
Solution: Handling different scale ranges - 105:24
Solution: Handling different scale ranges - 205:51
Solution: Handling different scale ranges - 305:56
Solution: Handling different scale ranges - 406:05
Solution: Handling different scale ranges - 506:10
Solution: Handling different scale ranges - 606:15
Solution: Handling different scale ranges - 706:18
Solution: Handling different scale ranges - 806:24
Solution: Handling different scale ranges - 906:25
Solution: Handling different scale ranges - 1006:31
Solution: Handling different scale ranges - 1106:54
Solution: Handling different scale ranges - 1207:07
Solution: Handling different scale ranges - 1307:17
Solution: Handling different scale ranges - 1407:19
Previous evaluation07:25
New evaluation + new annotations07:39
Before - 108:06
After - 108:13
Before - 208:26
After - 208:32
Before - 308:41
After - 308:42
Baselines08:43
Some methods are based on Viola&Jones08:54
Some methods are based on DPM08:59
Baselines are trained using AFLW09:06
DPM v5 baseline09:20
DPM on AFW dataset - 109:42
DPM on AFW dataset - 210:20
NMS threshold 0.510:24
NMS threshold 0.310:31
NMS matters a lot!10:42
Overlapping DPM detections - 110:49
Overlapping DPM detections - 211:21
Viola&Jones baseline11:25
Ours SquaresChnFtrs-511:44
What makes a face detector (truly) tick?11:53
Number of templates12:12
Results on Pascal Faces - 112:37
Results on Pascal Faces - 212:47
Results on Pascal Faces - 313:01
Results on Pascal Faces - 413:07
Results on Pascal Faces - 513:24
Results on AFW13:28
Results on FDDB13:42
Contributions14:13
Take home message14:34
Questions?14:51