Dense Captions

Overview

This capability provides requirements for the detailed, human-readable descriptions of 3D assets and comprised sub-hierarchies. These detailed captions may be used to support AI training and scene understanding.

Granularity

At a mininum, the root prim of the asset should contain documentation metadata, holding a Dense Caption string. Prims representing sub-objects or parts within the asset may additionally add dense captions.

Schema

Dense captions currently do not use a schema and they are conventional.

NVIDIA convention is to use the OpenUSD documentation (doc) string as shown below.

USDA Sample

def XForm "SportCar" (
    doc = "This is red, two door sports car with a tan, leather interior and chrome wheels. The driver's window is up and the headlights are off."  
)
{
    def Mesh "ConvertibleTop"
    {
        doc = "This is a retractable hard top."
    }
}

Requirements

Tags	Summary	Compatibility	Validator
🔑	The Root Prim must inlcude documentation metadata that describes the 3D Asset.	core usd
💎	Prims representing sub-objects or parts within the asset should be documented with additional dense captions.	core usd