0.25
0.5
0.75
1.25
1.5
1.75
2
Learning to Compare
Published on Sep 13, 20157185 Views
Related categories
Chapter list
Learning to Compare00:00
Overview: this talk - 101:01
Overview: this talk - 201:45
Learning similarity02:14
The setup - 104:57
The setup - 205:15
The setup - 305:29
The setup - 405:40
One motivation: nearest neighbour methods - 207:43
Outline - 108:18
Outline - 208:50
Outline - 309:27
Unsupervised approach - 109:41
Unsupervised approach - 210:20
Unsupervised approach - 311:26
Constrained Poisson model12:19
Deep auto-encoders - 113:53
Deep auto-encoders - 214:16
Deep auto-encoders - 315:06
Deep auto-encoders - 415:48
Extremely fast retrieval18:14
Hashing longer codes 219:59
Multi-index hashing23:12
Learning embeddings with a Siamese network - 126:48
Learning embeddings with a Siamese network - 227:31
Not a new idea!27:48
Convnets: single stage28:57
Convnets: typical architecture29:20
Embedding with a Siamese convnet29:34
Training Siamese nets31:30
Neighbourhood components analysis (NCA) - 132:07
Neighbourhood components analysis (NCA) - 232:34
Stochastic nearest neighbour - 134:08
Stochastic nearest neighbour - 237:21
NCA: loss38:06
Linear NCA: embeddings38:53
NCA: MNIST39:43
Nonlinear NCA40:06
Learning nonlinear NCA42:30
Limitations of NCA42:40
Class-conditional metric learning - 142:45
One motivation: nearest neighbour methods - 143:06
Class-conditional metric learning - 243:23
Class-conditional metric learning - 343:53
DrLIM (Dimensionality reduction by learning an invariant mapping)43:56
Spring analogy - 145:41
Spring analogy - 246:30
Triplet-based embedding48:58
Learning fine-grained image similarity with deep ranking51:00
How to: triplet sampling - 152:54
How to: triplet sampling - 254:05
Finding similarity data - 158:57
Finding similarity data - 259:16
Finding similarity data - 359:28
Hands by hand - 101:00:12
Pose-sensitive embeddings01:01:39
NCA regression01:02:04
Snowbird dataset01:02:58
Comparison of Approaches - 101:03:12
Comparison of Approaches - 201:03:48
Comparison of Approaches - 301:03:49
Results (qualitative)01:03:50
Results (quantitative)01:04:38
MPII Human Pose01:05:14
Pose embeddings01:06:39
Can we avoid explicit labeling of body parts?01:07:22
Weakly-supervised embeddings - 101:07:37
Weakly-supervised embeddings - 201:07:45
Weakly-supervised embeddings - 301:08:12
Zero-shot learning01:09:19
Summary01:12:13
Where to go from here?01:13:10