ClassifierPipeline » History » Version 1
J. Moringen, 08/30/2012 04:01 PM
initial version
| 1 | 1 | J. Moringen | h1. Classifier Pipeline |
|---|---|---|---|
| 2 | 1 | J. Moringen | |
| 3 | 1 | J. Moringen | h2. Structuring Ideas |
| 4 | 1 | J. Moringen | |
| 5 | 1 | J. Moringen | * Data-flow layer + control layer |
| 6 | 1 | J. Moringen | ** E.g. central controller manages component states according to some protocol/rules |
| 7 | 1 | J. Moringen | |
| 8 | 1 | J. Moringen | * Alternative: Subgraph lifecycles |
| 9 | 1 | J. Moringen | ** E.g. learning subgraph between data buffer and learner) |
| 10 | 1 | J. Moringen | |
| 11 | 1 | J. Moringen | h2. Training |
| 12 | 1 | J. Moringen | |
| 13 | 1 | J. Moringen | * Sequential processing: data source, keypoint, descriptor, feature |
| 14 | 1 | J. Moringen | * Processing Strategies (in CCA speech) |
| 15 | 1 | J. Moringen | ** First possibility |
| 16 | 1 | J. Moringen | *** As fast as possible |
| 17 | 1 | J. Moringen | *** Flow control is necessary |
| 18 | 1 | J. Moringen | *** Usually, all inputs are processed |
| 19 | 1 | J. Moringen | ** Second possibility |
| 20 | 1 | J. Moringen | *** "Live" data source (i.e. camera) |
| 21 | 1 | J. Moringen | * Multiple phases |
| 22 | 1 | J. Moringen | ** Collection |
| 23 | 1 | J. Moringen | ** Actual training |
| 24 | 1 | J. Moringen | ** Model is transmitted from learner to classifier |
| 25 | 1 | J. Moringen | |
| 26 | 1 | J. Moringen | h2. Classification |
| 27 | 1 | J. Moringen | |
| 28 | 1 | J. Moringen | * Static, linear data flow: source, keypoint, descriptor, feature, classifier |
| 29 | 1 | J. Moringen | * No buffering/persistence required |