Inherits Object.

Plugin base class for converting images into feature vectors

See:

• Constructors
• Static Functions
• Functions

Representations are used to convert images to feature vectors lazily (only when necessary). They are similar to Transforms in many respects but differ in a few key areas. Transforms should be used to construct feature vectors if it is desirable to construct a vector before evaluation that encompasses the entire feature space (or a smaller subset learned during training). Representations should be used if their is a large possible feature space but a few select features are necessary for a particular computation. This is often the case in tree architectures, where each node has an associated feature. The possible feature space is all of the features associated with all of the nodes, but the required features are only the features associated with nodes that are actually visited. The purpose of Representations is to allow these features to be calculated as needed instead of calculating all of the features before hand, which is less efficient.