Houdini 21.0 Machine Learning

Example Representation

The example-based ML nodes use packed primitives to represent examples. A geometry consisting entirely of packed primitives represents a collection of examples. This is convenient for processing collections of examples and multiple collections of examples can be joined together by merging their geometries. Individual examples can be extracted from a collection of examples by primitive index. The packed primitive representation helps avoid unnecessary copying of geometry.

Types of Examples

There are two types of examples: unlabeled examples and labeled examples. An unlabeled example contains a single geometry. A labeled example contains two geometries: an input component and a target component.

An unlabeled example is formed by using packing a geometry using Pack or ML Example with only the first input connected to a geometry.

A labeled example is formed by using ML Example, where both inputs are connected. A geometry containing labeled examples must have a primitive group called labeledexamples and all labeled-example primitives must be members of that group. Any packed primitive that is not part of a geometry or a member of the labeledexamples primitive group is treated as an unlabeled example, even if it contains two packed primitives.

Machine Learning

General Support

Supervised ML pipeline tools

ML Recipes

Animation and character-specific ML tools

Volume-specific ML tools

Image-specific ML tools

Reference