en
0.25
0.5
0.75
1.25
1.5
1.75
2
From buildings to cities: large scale visual search and mining in geotagged image collections
Published on Nov 26, 20122391 Views
Related categories
Chapter list
From buildings to cities: large scale visual search and mining in geo-tagged image collections00:00
Visual place recognition: visual search for places00:29
Availability of visual data01:24
Review: visual place recognition as particular object retrieval02:25
Success of text retrieval02:49
Particular object retrieval03:37
Example04:11
Results - 104:15
Results - 204:21
Matching of the local descriptors04:27
Mobile visual search04:39
Place recognition – the visual data is "structured"05:04
What are the features that characterize a location?05:47
Correctly recognized examples07:25
More correctly recognized examples07:47
What visual elements characterize a certain geo-spatial area?08:14
Where are these images from?08:56
How do they know?09:32
The goal10:05
Our hypothesis10:21
Street View Google Maps images - 110:40
Street View Google Maps images - 210:49
Sample images form Paris - 110:56
Sample images form Paris - 211:08
Sample images form Paris - 311:10
Mining object instances11:25
Mining visual patterns11:54
Representing visual patterns12:13
Hundreds of millions patches - 112:49
Hundreds of millions patches - 213:03
K-means clustering13:27
Not geo-informative13:40
Visually incoherent14:01
Our approach - 114:17
Our approach - 214:25
Our approach - 314:45
Learn patch similarity15:23
Paris: a few top elements16:00
Apply the same algorithm to different cities16:37
In the U.S.17:14
Applications - 117:32
Applications - 217:50
Applications - 318:11
Applications - 418:21
Applications - 518:30
Applications - 618:34
Applications - 718:40
Correspondence - 119:14
Correspondence - 219:16
Correspondence - 319:19
Correspondence - 419:24
Correspondence - 519:26
Conclusions and outlook19:27
"The dark matter of the Internet"20:11
New representations for visual data are needed20:38
Some of our recent work on representing video21:49
Acknowledgments22:24