en-de
en-fr
en-es
en-sl
en
en-zh
en-pt
0.25
0.5
0.75
1.25
1.5
1.75
2
Where machine vision needs help from machine learning
Published on 2011-08-0210566 Views
I'll describe where computer vision needs advances from computer science and machine learning. This talk will cover where computer vision works well: finding cars and faces, operating in controlled e
Related categories
Presentation
Where machine vision needs help from machine learning00:00
Outline - 100:06
The Taiyuan University of Technology Computer Center staff, and me (1987) - 100:40
The Taiyuan University of Technology Computer Center staff, and me (1987) - 200:57
Me and my wife, riding from the Foreigners’ Cafeteria01:02
Me in my office at the Computer Center01:10
VISION01:23
Goal of computer vision01:35
Some particular goals of computer vision02:02
Companies and applications02:42
COGNEX02:57
POSEIDON03:26
Natatorium03:41
Saved by a computer lifeguard03:44
Mobil Eye03:58
1998 Journal publication04:42
Sony EyeToy, 200305:01
Microsoft Kinect, 201005:13
identix06:04
Google Image Search, Google Goggles - 106:20
Google Image Search, Google Goggles - 206:33
The Computer Vision Industry07:07
Games and Gesture Recognition07:17
Industrial automation and inspection07:21
Object Recognition for Mobile Devices07:24
Three-dimensional modeling07:52
Outline - 209:16
Outline - 309:30
What makes computer vision hard? - 109:33
What makes computer vision hard? - 209:39
What makes computer vision hard? - 309:47
intra - class variation10:12
Object recognition issues10:36
Computer vision features over time11:21
What everyone looked like back then11:32
Features11:40
Objects11:46
Computer vision research results, 198611:57
Computer vision research results, 199212:37
Back to the present ...12:57
What has allowed us to make progress? - 113:15
What has allowed us to make progress? - 214:22
CVPR 2003 Tutorial - 114:29
CVPR 2003 Tutorial - 214:49
Invariant Local Features15:09
Hand under two different lighting conditions16:27
SIFT vector formation - 117:06
SIFT vector formation - 217:30
Feature stability to noise19:04
Feature stability to affine change19:38
Distinctiveness of features20:00
Figure 1220:34
Figure 1321:15
Building a Panorama - 121:39
Building a Panorama - 221:45
These feature point detectors and descriptors are the most important recent advance in computer vision and graphics21:56
S5ll another use for SIFT features22:28
Extarcting words22:50
Visual words23:16
Object recognition using visual words23:24
Now this starts to look like a learning theory problem24:15
What else do we want, to make progress?24:47
My poll of the top researchers in computer vision24:59
“How do you think computer science can best help computer vision?” - 125:21
“How do you think computer science can best help computer vision?” - 225:27
“How do you think computer science can best help computer vision?” - 325:36
Nearest neighbor search in high dimensions25:51
Fast Approximate Nearest Neighbors With Automatic Algorithm Configuration26:25
Comparison of different algorithms26:34
Additional structure present in NN problems for computer vision27:30
Another NN search problem, with structure: non-local means denoising28:31
Non-local means denoising algorithm28:52
An approx nearest-neighbor algo. that takes image spatial structure into account - 130:03
An approx nearest-neighbor algo. that takes image spatial structure into account - 230:20
An approx nearest-neighbor algo. that takes image spatial structure into account - 330:26
An approx nearest-neighbor algo. that takes image spatial structure into account - 430:36
An approx nearest-neighbor algo. that takes image spatial structure into account - 530:48
An approx nearest-neighbor algo. that takes image spatial structure into account - 630:58
An approx nearest-neighbor algo. that takes image spatial structure into account - 731:02
An approx nearest-neighbor algo. that takes image spatial structure into account - 831:30
An approx nearest-neighbor algo. that takes image spatial structure into account - 931:44
Problem - 134:11
Another commonly expressed need: help in scaling up algorithms34:27
Problem - 235:06
Outline - 435:11
Priors on images35:30
Removing camera shake - 136:42
Removing camera shake - 237:02
Close-up - 137:17
Close-up - 237:18
Close-up - 337:20
Image formation process37:29
Multiple possible solutions - 138:04
Multiple possible solutions - 238:08
Multiple possible solutions - 338:23
Multiple possible solutions - 438:30
Multiple possible solutions - 538:49
Is each of the images that follow sharp or blurred? 38:51
Picture - 138:57
Picture - 239:02
Picture - 339:04
Natural image statistics39:19
Blury images have different statistics40:05
Parametric distribution40:18
Three sources of information - 140:26
Three sources of information - 240:28
Three sources of information - 340:34
Three sources of information - 440:39
Bayesian estimate of latent image, x, and blur kernel, b.40:45
Original photograph41:23
Our output41:32
Matlab's deconvblind41:43
Close-up of garland41:48
A stronger prior might help us deblur this image42:06
Problem - 342:22
Texture Synthesis by Non-parametric Sampling42:30
Algorithm43:15
Picture - 443:27
Picture - 543:34
Picture - 643:38
Problem - 443:53
Special case of an image prior44:14
Some methods of approximate inference in MRF’s45:28
MRF/CRF wishes - 145:55
Input image46:10
MRF/CRF wishes - 246:57
Image Segmentation with Bounding Box Prior47:31
Other constraints48:37
Problem - 548:46
Outline - 549:01
Compressed sensing - 149:06
Compressed sensing - 249:15
Problem - 649:33
Large, noisy datasets49:45
Problem - 750:48
Shai Avidan50:51
Blind vision50:55
Problem - 850:57
me51:55
Continuous to discrete representations51:58
Problem - 952:52
Deva Ramanan52:59
Evaluate easily over a powerset of all segmentations53:01
Problem - 1053:24
Alyosha Efros53:28
Efros and Hoiem comments53:31
Really, we’d like another breakthrough...54:02
David Lowe - 154:18
David Lowe - 254:23
Most references are in citation list of this manuscript54:45
Computer vision academic culture54:57
A computer graphics application of nearest ‐ neighbor finding in high dimensions - 156:12
A computer graphics application of nearest ‐ neighbor finding in high dimensions - 256:17
The image database56:40
Obtaining semantically coherent themes56:46
Image representation57:15
Basic camera motions57:40
Scene matching with camera view transformations: Translation - 157:41
Scene matching with camera view transformations: Translation - 257:48
Scene matching with camera view transformations: Translation - 358:03
Scene matching with camera view transformations: Translation - 458:14
Scene matching with camera view transformations: Translation - 558:20
Scene matching with camera view transformations: Translation - 658:29
Scene matching with camera view transformations: Camera rotation - 158:55
Scene matching with camera view transformations: Camera rotation - 259:03
Scene matching with camera view transformations: Camera rotation - 359:07
Scene matching with camera view transformations: Camera rotation - 459:08
Scene matching with camera view transformations: Camera rotation - 559:09
More “infinite” images – camera translation59:11
Video - 159:19
Video - 259:47
Video - 301:00:08