en
0.25
0.5
0.75
1.25
1.5
1.75
2
The quest for a theory of vision: from the level framework (revised) to the invariances of the ventral stream
Published on Nov 12, 20126095 Views
Related categories
Chapter list
The quest for a theory of vision: from the level framework (revised) to the invariances of the ventral stream00:00
Collaborators in recent work04:03
Vision as Intelligence04:27
Vision: a very difficult computational problem05:34
The problem of intelligence (in particular, vision)05:59
Deep Blue06:33
Google06:42
Kinect06:51
Watson06:56
Siri07:05
Drones07:13
Vision @CBCL, ~20 years ago - 107:57
Vision @CBCL, ~20 years ago - 208:04
Vision @CBCL, ~20 years ago - 308:15
Vision @CBCL, ~18 years ago - 108:24
Vision @CBCL, ~18 years ago - 208:34
Vision @CBCL, ~18 years ago - 308:35
Vision @CBCL, ~18 years ago - 408:38
Vision, ~ now08:54
MobilEye - 108:59
MobilEye - 209:10
Advertisement by Volvo09:31
Recent successes in ML and AI: computer vision10:22
A "fullTuring" test for vision? - 110:49
A "fullTuring" test for vision? - 210:56
A "fullTuring" test for vision? - 311:04
A "fullTuring" test for vision? - 411:08
A "fullTuring" test for vision? - 511:11
A "fullTuring" test for vision? - 611:13
A "fullTuring" test for vision? - 711:14
A "fullTuring" test for vision? - 811:18
A "fullTuring" test for vision? - 911:20
A "fullTuring" test for vision? - 1011:41
The next big step: brain12:35
Vision in the Brain13:03
Visual Object Recognition: the ventral stream (macaque)13:46
Recognition in the Ventral Stream: "standard" feedforward model15:59
Recognition in Visual Cortex: "classical model", selective and invariant17:00
Model "works": it accounts for physiology17:35
Model "works": it accounts for psychophysics - 118:08
Model "works": it accounts for psychophysics - 218:09
Model "works": it accounts for psychophysics - 318:33
Model "works": it performs well at computational level - 119:14
Model "works": it performs well at computational level - 219:16
Model "works": it performs well at computational level - 319:35
Model "works": it performs well at computational level - 419:47
Model "works": it performs well at computational level - 519:49
Visual Cortex: models and theories20:28
A theory (unpublished) of the ventral stream: too nice to be true?21:36
Motivation22:17
Motivation: transformations may be a main difficulty for (biological) object recognition23:36
Motivation: transformations may be the main difficulty for (biological) object recognition24:15
Conjecture: the key computational problem "solved" by the ventral stream24:53
A theory of invariant recognition in the ventral stream25:17
Some of the questions answered by the theory26:00
Key theorems27:21
Gabor-like tuning with "universal constants" in simple cells27:48
2 Different stages in the theory29:33
Image representation in the ventral stream30:43
Templates and signature31:42
A motivation for signatures: the Johnson-Lindenstrauss theorem (features do not matter much!)32:26
Geometric transformations33:37
Learning to be invariant for any new object33:59
Templatebook35:43
Invariance lemma36:01
Invariant Signature36:35
One-layer architectures: pro and contra36:54
Whole-parts invariance37:28
Whole-parts theorem38:26
Part II: Linking Conjecture38:40
Tuning and eigenvectors of covariance matrix40:19
Gaussian aperture: the cortical equation40:42
The cortical equation: general properties42:04
Cortical equation in 2D: natural images, Gabor-like receptive fields - 142:36
Cortical equation in 2D: natural images, Gabor-like receptive fields - 243:02
Too nice to be true!43:33
Beyond V1, towards V2 and V444:08
We are working on implementing the full theory44:41
Class-specific modules45:24
A theory of biological vision45:55