0.25
0.5
0.75
1.25
1.5
1.75
2
Presentation
Learning to Compare00:00
Overview: this talk - 117:02:18
Overview: this talk - 229:19:53
Learning similarity37:13:33
The setup - 182:34:48
The setup - 287:32:48
The setup - 391:26:38
The setup - 494:34:57
One motivation: nearest neighbour methods - 2128:41:14
Outline - 1138:30:52
Unsupervised approach - 1161:39:02
Unsupervised approach - 2172:17:20
Unsupervised approach - 3190:49:15
Constrained Poisson model205:24:18
Deep auto-encoders - 1231:27:11
Deep auto-encoders - 2237:51:42
Deep auto-encoders - 3251:55:31
Deep auto-encoders - 4263:20:37
Extremely fast retrieval303:57:44
Hashing longer codes 2333:06:04
Multi-index hashing386:46:34
Learning embeddings with a Siamese network - 1446:48:39
Learning embeddings with a Siamese network - 2458:48:41
Not a new idea!463:27:43
Convnets: single stage482:39:37
Convnets: typical architecture488:58:19
Embedding with a Siamese convnet492:50:00
Training Siamese nets525:00:57
Neighbourhood components analysis (NCA) - 1535:23:16
Neighbourhood components analysis (NCA) - 2542:50:48
Stochastic nearest neighbour - 1568:56:29
Stochastic nearest neighbour - 2622:40:31
NCA: loss635:06:19
Linear NCA: embeddings648:12:30
NCA: MNIST662:06:29
Nonlinear NCA668:35:00
Learning nonlinear NCA708:35:49
Limitations of NCA711:07:57
Class-conditional metric learning - 1712:40:21
One motivation: nearest neighbour methods - 1718:25:18
Class-conditional metric learning - 2723:12:21
Class-conditional metric learning - 3731:23:20
DrLIM (Dimensionality reduction by learning an invariant mapping)732:26:09
Spring analogy - 1761:35:55
Spring analogy - 2775:06:32
Triplet-based embedding816:21:52
Learning fine-grained image similarity with deep ranking850:06:05
How to: triplet sampling - 1881:54:34
How to: triplet sampling - 2901:37:51
Finding similarity data - 1982:39:44
Finding similarity data - 2987:53:29
Finding similarity data - 3991:14:29
Hands by hand - 11003:26:30
Pose-sensitive embeddings1027:42:54
NCA regression1034:41:04
Snowbird dataset1049:33:45
Comparison of Approaches - 11053:34:54
Comparison of Approaches - 21063:33:43
Comparison of Approaches - 31063:44:21
Results (qualitative)1064:06:20
Results (quantitative)1077:17:35
MPII Human Pose1087:14:50
Pose embeddings1111:00:12
Can we avoid explicit labeling of body parts?1122:49:04
Weakly-supervised embeddings - 11127:07:17
Weakly-supervised embeddings - 21129:15:38
Weakly-supervised embeddings - 31136:44:34
Zero-shot learning1155:26:20
Summary1203:42:10
Where to go from here?1219:36:14