en
0.25
0.5
0.75
1.25
1.5
1.75
2
“Clustering by Composition” for Unsupervised Discovery of Image Categories
Published on Nov 12, 20124179 Views
We define a good image cluster as one in which images can be easily composed (like a puzzle) using pieces from each other, while are difficult to compose from images outside the cluster. The larger a
Related categories
Chapter list
“Clustering by Composition”: Unsupervised Discovery of Image Categories00:00
Unsupervised Discovery of Image Categories (1)00:11
Unsupervised Discovery of Image Categories (2)00:14
Goal: Separate into 2 clusters (Yoga & Ballet) (3)00:31
Goal: Separate into 2 clusters (Yoga & Ballet) (4)01:13
Goal: Separate into 2 clusters (Yoga & Ballet) (5)02:35
Goal: Separate into 2 clusters (Yoga & Ballet) (6)02:46
“Affinity by Composition” (1)03:05
“Affinity by Composition” (2)03:36
“Affinity by Composition” (3)03:48
“Affinity by Composition” (4)03:53
“Affinity by Composition” (5)04:07
Statistically Significant Descriptors (1)04:14
Statistically Significant Descriptors (2)05:01
Statistically Significant Descriptors (3)05:10
Statistically Significant Descriptors (4)05:19
“Affinity by Composition” (6)06:05
“Affinity by Composition” (7)06:16
“Affinity by Composition” (8)06:22
“Affinity by Composition” (9)06:28
“Affinity by Composition” (10)06:37
“Affinity by Composition” (11)06:41
“Affinity by Composition” (12)06:45
“Region Match” (1)06:47
“Region Match” (2)06:55
“Region Match” (3)07:02
“Region Match” (4)07:07
“Region Match” (5)07:26
“Region Match” (6)07:46
“Region Match” (7)08:14
Going back to our Clustering problem… (1)08:26
Going back to our Clustering problem… (2)09:00
Going back to our Clustering problem… (3)09:19
Going back to our Clustering problem… (4)09:50
Going back to our Clustering problem… (5)10:08
Our Full Clustering Algorithm10:17
Experiments11:09
Experiments on Tiny Dataset (1)11:48
Experiments on Tiny Dataset (2)12:19
Experiments on Tiny Dataset (3)12:28
Experiments on Tiny Dataset (4)12:36
PASCAL subset (4 classes) -112:52
PASCAL subset (4 classes) -213:13
PASCAL subset (4 classes) -313:35
Thank you!13:55