schema.usda

#usda 1.0
(
    "This file describes the USD Geometric schemata for code generation."
    subLayers = [
        @usd/schema.usda@
    ]
)

over "GLOBAL" (
    customData = {
        string libraryName           = "usdGeom"
        string libraryPath           = "pxr/usd/usdGeom"
        # string libraryPrefix       = "UsdGeom"
        # string tokensPrefix        = "UsdGeom"
        dictionary libraryTokens = {
            dictionary interpolation = {
                string doc = """UsdGeomPrimvar - How a Primvar interpolates
                across a primitive; equivalent to RenderMan's \\ref Usd_InterpolationVals "class specifier" """
            }
            dictionary elementSize = {
                string doc = """UsdGeomPrimvar - The number of values in the
                value array that must be aggregated for each element on the 
                primitive."""
            }
            dictionary unauthoredValuesIndex = {
                string doc = """UsdGeomPrimvar - The index that represents 
                unauthored values in the indices array of an indexed primvar."""
            }
            dictionary constant ={
                string doc = """Possible value for UsdGeomPrimvar::SetInterpolation.
                Default value for UsdGeomPrimvar::GetInterpolation. One value
                remains constant over the entire surface primitive."""
            }
            dictionary uniform = {
                string doc = """Possible value for UsdGeomPrimvar::SetInterpolation.
                One value remains constant for each uv patch segment of the
                surface primitive (which is a \\em face for meshes)."""
            }
            dictionary varying = {
                string doc = """Possible value for UsdGeomPrimvar::SetInterpolation.
                Four values are interpolated over each uv patch segment of the 
                surface. Bilinear interpolation is used for interpolation 
                between the four values."""
            }
            dictionary vertex = {
                string doc = """Possible value for UsdGeomPrimvar::SetInterpolation.
                Values are interpolated between each vertex in the surface
                primitive. The basis function of the surface is used for 
                interpolation between vertices."""
            }
            dictionary faceVarying = {
                string doc = """Possible value for UsdGeomPrimvar::SetInterpolation.
                For polygons and subdivision surfaces, four values are
                interpolated over each face of the mesh. Bilinear interpolation 
                is used for interpolation between the four values."""
            }
            dictionary extentsHint = {
                string doc = """Name of the attribute used to author extents
                hints at the root of leaf models. Extents hints are stored by purpose
                as a vector of GfVec3f values. They are ordered based on the order
                of purpose tokens returned by 
                UsdGeomImageable::GetOrderedPurposeTokens."""
            }
            dictionary upAxis = {
                string doc = """Stage-level metadata that encodes a scene's
                orientation as a token whose value can be "Y" or "Z"."""
            }
            dictionary metersPerUnit = {
                string doc = """Stage-level metadata that encodes a scene's
                linear unit of measure as meters per encoded unit."""
            }
            dictionary inactiveIds = {
                string doc = """int64listop prim metadata that specifies
                the PointInstancer ids that should be masked (unrenderable)
                over all time."""
            }
            dictionary partition = {
                string doc = """A type of family of GeomSubsets. It implies 
                that every element appears exacly once in only one of the 
                subsets in the family."""
            }
            dictionary nonOverlapping = {
                string doc = """A type of family of GeomSubsets. It implies that 
                the elements in the various subsets belonging to the family are 
                mutually exclusive, i.e., an element that appears in one 
                subset may not belong to any other subset in the family."""
            }
            dictionary unrestricted = {
                string doc = """A type of family of GeomSubsets. It implies that
                there are no restrictions w.r.t. the membership of elements in 
                the subsets. There could be overlapping members in subsets 
                belonging to the family and the union of all subsets in the 
                family may not contain all the elements."""
            }
        }
    }
)
{
}

class "Imageable" (
    inherits = </Typed>
    doc = """Base class for all prims that may require rendering or 
    visualization of some sort. The primary attributes of Imageable 
    are \\em visibility and \\em purpose, which each provide instructions for
    what geometry should be included for processing by rendering and other
    computations.

    \\deprecated Imageable also provides API for accessing primvars, which
    has been moved to the UsdGeomPrimvarsAPI schema, because primvars can now
    be applied on non-Imageable prim types.  This API is planned
    to be removed, UsdGeomPrimvarsAPI should be used directly instead."""
    customData = {
        string extraIncludes = """
#include "pxr/base/gf/bbox3d.h"
#include "pxr/usd/usdGeom/primvar.h" """
    }
) {
    token visibility = "inherited" (
        allowedTokens = ["inherited", "invisible"]
        doc = """Visibility is meant to be the simplest form of "pruning" 
        visibility that is supported by most DCC apps.  Visibility is 
        animatable, allowing a sub-tree of geometry to be present for some 
        segment of a shot, and absent from others; unlike the action of 
        deactivating geometry prims, invisible geometry is still 
        available for inspection, for positioning, for defining volumes, etc."""
    )
    
    uniform token purpose = "default" (
        allowedTokens = ["default", "render", "proxy", "guide"]
        doc = """Purpose is a concept we have found useful in our pipeline for 
        classifying geometry into categories that can each be independently
        included or excluded from traversals of prims on a stage, such as
        rendering or bounding-box computation traversals.  The fallback
        purpose, \\em default indicates that a prim has "no special purpose"
        and should generally be included in all traversals.  Subtrees rooted
        at a prim with purpose \\em render should generally only be included
        when performing a "final quality" render.  Subtrees rooted at a prim
        with purpose \\em proxy should generally only be included when 
        performing a lightweight proxy render (such as openGL).  Finally,
        subtrees rooted at a prim with purpose \\em guide should generally
        only be included when an interactive application has been explicitly
        asked to "show guides". 
        
        In the previous paragraph, when we say "subtrees rooted at a prim",
        we mean the most ancestral or tallest subtree that has an authored,
        non-default opinion.  If the purpose of </RootPrim> is set to 
        "render", then the effective purpose of </RootPrim/ChildPrim> will
        be "render" even if that prim has a different authored value for
        purpose.  <b>See ComputePurpose() for details of how purpose 
        inherits down namespace</b>.
        
        As demonstrated in UsdGeomBBoxCache, a traverser should be ready to 
        accept combinations of included purposes as an input.
        
        Purpose \\em render can be useful in creating "light blocker"
        geometry for raytracing interior scenes.  Purposes \\em render and
        \\em proxy can be used together to partition a complicated model
        into a lightweight proxy representation for interactive use, and a
        fully realized, potentially quite heavy, representation for rendering.
        One can use UsdVariantSets to create proxy representations, but doing
        so requires that we recompose parts of the UsdStage in order to change
        to a different runtime level of detail, and that does not interact
        well with the needs of multithreaded rendering. Purpose provides us with
        a better tool for dynamic, interactive complexity management."""
    )
    rel proxyPrim (
        doc = """The \\em proxyPrim relationship allows us to link a
        prim whose \\em purpose is "render" to its (single target)
        purpose="proxy" prim.  This is entirely optional, but can be
        useful in several scenarios:
        
        \\li In a pipeline that does pruning (for complexity management)
        by deactivating prims composed from asset references, when we
        deactivate a purpose="render" prim, we will be able to discover
        and additionally deactivate its associated purpose="proxy" prim,
        so that preview renders reflect the pruning accurately.
        
        \\li DCC importers may be able to make more aggressive optimizations
        for interactive processing and display if they can discover the proxy
        for a given render prim.
        
        \\li With a little more work, a Hydra-based application will be able
        to map a picked proxy prim back to its render geometry for selection.

        \\note It is only valid to author the proxyPrim relationship on
        prims whose purpose is "render"."""
    )
}

class "PrimvarsAPI" (
    inherits = </APISchemaBase>
    doc = """UsdGeomPrimvarsAPI encodes geometric "primitive variables",
    as UsdGeomPrimvar, which interpolate across a primitive's topology,
    can override shader inputs, and inherit down namespace.
    
    \\section usdGeom_PrimvarFetchingAPI Which Method to Use to Retrieve Primvars
     
     While creating primvars is unambiguous (CreatePrimvar()), there are quite
     a few methods available for retrieving primvars, making it potentially
     confusing knowing which one to use.  Here are some guidelines:
     
     \\li If you are populating a GUI with the primvars already available for 
     authoring values on a prim, use GetPrimvars().
     \\li If you want all of the "useful" (e.g. to a renderer) primvars
     available at a prim, including those inherited from ancestor prims, use
     FindPrimvarsWithInheritance().  Note that doing so individually for many
     prims will be inefficient.
     \\li To find a particular primvar defined directly on a prim, which may
     or may not provide a value, use GetPrimvar().
     \\li To find a particular primvar defined on a prim or inherited from
     ancestors, which may or may not provide a value, use 
     FindPrimvarWithInheritance().
     \\li To *efficiently* query for primvars using the overloads of
     FindPrimvarWithInheritance() and FindPrimvarsWithInheritance(), one
     must first cache the results of FindIncrementallyInheritablePrimvars() for
     each non-leaf prim on the stage. """ 
    customData = {
        token apiSchemaType = "nonApplied"
        string extraIncludes = """
#include "pxr/usd/usdGeom/primvar.h" """
    }
) {
}

class "Xformable" (
    inherits = </Imageable>
    customData = {
        string extraIncludes = """
#include "pxr/usd/usdGeom/xformOp.h" 
#include <vector> """
    }
    doc = """Base class for all transformable prims, which allows arbitrary
    sequences of component affine transformations to be encoded.

    \\note 
    You may find it useful to review \\ref UsdGeom_LinAlgBasics while reading
    this class description.
    
    <b>Supported Component Transformation Operations</b>
    
    UsdGeomXformable currently supports arbitrary sequences of the following
    operations, each of which can be encoded in an attribute of the proper
    shape in any supported precision:
    \\li translate - 3D
    \\li scale     - 3D
    \\li rotateX   - 1D angle in degrees
    \\li rotateY   - 1D angle in degrees
    \\li rotateZ   - 1D angle in degrees
    \\li rotateABC - 3D where ABC can be any combination of the six principle
                        Euler Angle sets: XYZ, XZY, YXZ, YZX, ZXY, ZYX.  See
                        \\ref usdGeom_rotationPackingOrder "note on rotation packing order"
    \\li orient    - 4D (quaternion)
    \\li transform - 4x4D 
    
    <b>Creating a Component Transformation</b>
    
    To add components to a UsdGeomXformable prim, simply call AddXformOp()
    with the desired op type, as enumerated in \\ref UsdGeomXformOp::Type,
    and the desired precision, which is one of \\ref UsdGeomXformOp::Precision.
    Optionally, you can also provide an "op suffix" for the operator that 
    disambiguates it from other components of the same type on the same prim.  
    Application-specific transform schemas can use the suffixes to fill a role 
    similar to that played by AbcGeom::XformOp's "Hint" enums for their own 
    round-tripping logic.
    
    We also provide specific "Add" API for each type, for clarity and 
    conciseness, e.g. AddTranslateOp(), AddRotateXYZOp() etc.
    
    AddXformOp() will return a UsdGeomXformOp object, which is a schema on a 
    newly created UsdAttribute that provides convenience API for authoring
    and computing the component transformations.  The UsdGeomXformOp can then
    be used to author any number of timesamples and default for the op.
    
    Each successive call to AddXformOp() adds an operator that will be applied
    "more locally" than the preceding operator, just as if we were pushing
    transforms onto a transformation stack - which is precisely what should
    happen when the operators are consumed by a reader.
    
    \\note
    If you can, please try to use the UsdGeomXformCommonAPI, which wraps
    the UsdGeomXformable with an interface in which Op creation is taken
    care of for you, and there is a much higher chance that the data you
    author will be importable without flattening into other DCC's, as it
    conforms to a fixed set of Scale-Rotate-Translate Ops.
    
    \\sa \\ref usdGeom_xformableExamples "Using the Authoring API"
    
    <b>Data Encoding and Op Ordering</b>
    
    Because there is no "fixed schema" of operations, all of the attributes
    that encode transform operations are dynamic, and are scoped in 
    the namespace "xformOp". The second component of an attribute's name provides
    the \\em type of operation, as listed above.  An "xformOp" attribute can 
    have additional namespace components derived from the \\em opSuffix argument 
    to the AddXformOp() suite of methods, which provides a preferred way of 
    naming the ops such that we can have multiple "translate" ops with unique
    attribute names. For example, in the attribute named 
    "xformOp:translate:maya:pivot", "translate" is the type of operation and
    "maya:pivot" is the suffix.
    
    The following ordered list of attribute declarations in usda
    define a basic Scale-Rotate-Translate with XYZ Euler angles, wherein the
    translation is double-precision, and the remainder of the ops are single,
    in which we will:

    <ol>
    <li> Scale by 2.0 in each dimension
    <li> Rotate about the X, Y, and Z axes by 30, 60, and 90 degrees, respectively
    <li> Translate by 100 units in the Y direction
    </ol>

    \\code
    float3 xformOp:rotateXYZ = (30, 60, 90)
    float3 xformOp:scale = (2, 2, 2)
    double3 xformOp:translate = (0, 100, 0)
    uniform token[] xformOpOrder = [ "xformOp:translate", "xformOp:rotateXYZ", "xformOp:scale" ]
    \\endcode

    The attributes appear in the dictionary order in which USD, by default,
    sorts them.  To ensure the ops are recovered and evaluated in the correct
    order, the schema introduces the **xformOpOrder** attribute, which
    contains the names of the op attributes, in the precise sequence in which
    they should be pushed onto a transform stack. **Note** that the order is
    opposite to what you might expect, given the matrix algebra described in
    \\ref UsdGeom_LinAlgBasics.  This also dictates order of op creation,
    since each call to AddXformOp() adds a new op to the end of the
    \\b xformOpOrder array, as a new "most-local" operation.  See 
    \\ref usdGeom_xformableExamples "Example 2 below" for C++ code that could
    have produced this USD.
    
    If it were important for the prim's rotations to be independently 
    overridable, we could equivalently (at some performance cost) encode
    the transformation also like so:
    \\code
    float xformOp:rotateX = 30
    float xformOp:rotateY = 60
    float xformOp:rotateZ = 90
    float3 xformOp:scale = (2, 2, 2)
    double3 xformOp:translate = (0, 100, 0)
    uniform token[] xformOpOrder = [ "xformOp:translate", "xformOp:rotateZ", "xformOp:rotateY", "xformOp:rotateX", "xformOp:scale" ]
    \\endcode
    
    Again, note that although we are encoding an XYZ rotation, the three
    rotations appear in the **xformOpOrder** in the opposite order, with Z,
    followed, by Y, followed by X.

    Were we to add a Maya-style scalePivot to the above example, it might 
    look like the following:
    \\code
    float3 xformOp:rotateXYZ = (30, 60, 90)
    float3 xformOp:scale = (2, 2, 2)
    double3 xformOp:translate = (0, 100, 0)
    double3 xformOp:translate:scalePivot
    uniform token[] xformOpOrder = [ "xformOp:translate", "xformOp:rotateXYZ", "xformOp:translate:scalePivot", "xformOp:scale" ]
    \\endcode

    <b>Paired "Inverted" Ops</b>

    We have been claiming that the ordered list of ops serves as a set
    of instructions to a transform stack, but you may have noticed in the last
    example that there is a missing operation - the pivot for the scale op
    needs to be applied in its inverse-form as a final (most local) op!  In the 
    AbcGeom::Xform schema, we would have encoded an actual "final" translation
    op whose value was authored by the exporter as the negation of the pivot's
    value.  However, doing so would be brittle in USD, given that each op can
    be independently overridden, and the constraint that one attribute must be
    maintained as the negation of the other in order for successful
    re-importation of the schema cannot be expressed in USD.
    
    Our solution leverages the **xformOpOrder** member of the schema, which,
    in addition to ordering the ops, may also contain one of two special
    tokens that address the paired op and "stack resetting" behavior.

    The "paired op" behavior is encoded as an "!invert!" prefix in 
    \\b xformOpOrder, as the result of an AddXformOp(isInverseOp=True) call.  
    The \\b xformOpOrder for the last example would look like:
    \\code
    uniform token[] xformOpOrder = [ "xformOp:translate", "xformOp:rotateXYZ", "xformOp:translate:scalePivot", "xformOp:scale", "!invert!xformOp:translate:scalePivot" ]
    \\endcode
    
    When asked for its value via UsdGeomXformOp::GetOpTransform(), an
    "inverted" Op (i.e. the "inverted" half of a set of paired Ops) will fetch 
    the value of its paired attribute and return its negation.  This works for 
    all op types - an error will be issued if a "transform" type op is singular 
    and cannot be inverted. When getting the authored value of an inverted op 
    via UsdGeomXformOp::Get(), the raw, uninverted value of the associated
    attribute is returned.

    For the sake of robustness, <b>setting a value on an inverted op is disallowed.</b>
    Attempting to set a value on an inverted op will result in a coding error 
    and no value being set. 
    
    <b>Resetting the Transform Stack</b>

    The other special op/token that can appear in \\em xformOpOrder is
    \\em "!resetXformStack!", which, appearing as the first element of 
    \\em xformOpOrder, indicates this prim should not inherit the transformation
    of its namespace parent.  See SetResetXformStack()

    <b>Expected Behavior for "Missing" Ops</b>
    
    If an importer expects Scale-Rotate-Translate operations, but a prim
    has only translate and rotate ops authored, the importer should assume
    an identity scale.  This allows us to optimize the data a bit, if only
    a few components of a very rich schema (like Maya's) are authored in the
    app.
    
    \\anchor usdGeom_xformableExamples
    <b>Using the C++ API</b>
    
    #1. Creating a simple transform matrix encoding
    \\snippet examples.cpp CreateMatrixWithDefault
    
    #2. Creating the simple SRT from the example above
    \\snippet examples.cpp CreateExampleSRT
    
    #3. Creating a parameterized SRT with pivot using UsdGeomXformCommonAPI
    \\snippet examples.cpp CreateSRTWithDefaults
    
    #4. Creating a rotate-only pivot transform with animated
    rotation and translation
    \\snippet examples.cpp CreateAnimatedTransform
    
"""
) {

    uniform token[] xformOpOrder (
        doc = """Encodes the sequence of transformation operations in the
        order in which they should be pushed onto a transform stack while
        visiting a UsdStage's prims in a graph traversal that will effect
        the desired positioning for this prim and its descendant prims.
        
        You should rarely, if ever, need to manipulate this attribute directly.
        It is managed by the AddXformOp(), SetResetXformStack(), and
        SetXformOpOrder(), and consulted by GetOrderedXformOps() and
        GetLocalTransformation()."""
    )
}

class Scope "Scope" (
    inherits = </Imageable>
    doc = """Scope is the simplest grouping primitive, and does not carry the
    baggage of transformability.  Note that transforms should inherit down
    through a Scope successfully - it is just a guaranteed no-op from a
    transformability perspective."""
) {
}

class Xform "Xform" (
    inherits = </Xformable>
    doc = """Concrete prim schema for a transform, which implements Xformable """
) {
}

class "Boundable" (
    inherits = </Xformable>
    doc = """Boundable introduces the ability for a prim to persistently
    cache a rectilinear, local-space, extent.
    
    \\section UsdGeom_Boundable_Extent Why Extent and not Bounds ?
    Boundable introduces the notion of "extent", which is a cached computation
    of a prim's local-space 3D range for its resolved attributes <b>at the
    layer and time in which extent is authored</b>.  We have found that with
    composed scene description, attempting to cache pre-computed bounds at
    interior prims in a scene graph is very fragile, given the ease with which
    one can author a single attribute in a stronger layer that can invalidate
    many authored caches - or with which a re-published, referenced asset can
    do the same.
    
    Therefore, we limit to precomputing (generally) leaf-prim extent, which
    avoids the need to read in large point arrays to compute bounds, and
    provides UsdGeomBBoxCache the means to efficiently compute and
    (session-only) cache intermediate bounds.  You are free to compute and
    author intermediate bounds into your scenes, of course, which may work
    well if you have sufficient locks on your pipeline to guarantee that once
    authored, the geometry and transforms upon which they are based will
    remain unchanged, or if accuracy of the bounds is not an ironclad
    requisite. 
    
    When intermediate bounds are authored on Boundable parents, the child prims
    will be pruned from BBox computation; the authored extent is expected to
    incorporate all child bounds."""  
)
{
    # XXX: Note this is really a GfRange3f, which is not fully supported
    # in Vt I/O.
    float3[] extent (
        doc = """Extent is a three dimensional range measuring the geometric
        extent of the authored gprim in its own local space (i.e. its own
        transform not applied), \\em without accounting for any shader-induced
        displacement.  Whenever any geometry-affecting attribute is authored
        for any gprim in a layer, extent must also be authored at the same
        timesample; failure to do so will result in incorrect bounds-computation.
        \\sa \\ref UsdGeom_Boundable_Extent.
        
        An authored extent on a prim which has children is expected to include
        the extent of all children, as they will be pruned from BBox computation
        during traversal."""
    )
}

class "Gprim" ( 
    inherits = </Boundable> 
    doc = """Base class for all geometric primitives.  
    
    Gprim encodes basic graphical properties such as \\em doubleSided and
    \\em orientation, and provides primvars for "display color" and "display
    opacity" that travel with geometry to be used as shader overrides.  """

) {
    color3f[] primvars:displayColor (
        customData = {
            string apiName = "displayColor"
        }
        doc = """It is useful to have an "official" colorSet that can be used
        as a display or modeling color, even in the absence of any specified
        shader for a gprim.  DisplayColor serves this role; because it is a
        UsdGeomPrimvar, it can also be used as a gprim override for any shader
        that consumes a \\em displayColor parameter."""
    )
    
    float[] primvars:displayOpacity (
        customData = {
            string apiName = "displayOpacity"
        }
        doc = """Companion to \\em displayColor that specifies opacity, broken
        out as an independent attribute rather than an rgba color, both so that
        each can be independently overridden, and because shaders rarely consume
        rgba parameters."""
    )
    
    uniform bool doubleSided = false (
        doc = """Although some renderers treat all parametric or polygonal
        surfaces as if they were effectively laminae with outward-facing
        normals on both sides, some renderers derive significant optimizations
        by considering these surfaces to have only a single outward side,
        typically determined by control-point winding order and/or 
        \\em orientation.  By doing so they can perform "backface culling" to
        avoid drawing the many polygons of most closed surfaces that face away
        from the viewer.
        
        However, it is often advantageous to model thin objects such as paper
        and cloth as single, open surfaces that must be viewable from both
        sides, always.  Setting a gprim's \\em doubleSided attribute to 
        \\c true instructs all renderers to disable optimizations such as
        backface culling for the gprim, and attempt (not all renderers are able
        to do so, but the USD reference GL renderer always will) to provide
        forward-facing normals on each side of the surface for lighting
        calculations."""
    )
    
    uniform token orientation = "rightHanded" (
        allowedTokens = ["rightHanded", "leftHanded"]
        doc = """Orientation specifies whether the gprim's surface normal 
        should be computed using the right hand rule, or the left hand rule.
        Please see \\ref UsdGeom_WindingOrder for a deeper explanation and
        generalization of orientation to composed scenes with transformation
        hierarchies."""
        )
}

class Cube "Cube" (
    inherits = </Gprim>
    doc = """Defines a primitive rectilinear cube centered at the origin.
    
    The fallback values for Cube, Sphere, Cone, and Cylinder are set so that
    they all pack into the same volume/bounds."""
    customData = {
        dictionary extraPlugInfo = {
            bool implementsComputeExtent = true
        }
    }
) {
    double size = 2.0 (
        doc = """Indicates the length of each edge of the cube.  If you
        author \\em size you must also author \\em extent.
        
        \\sa GetExtentAttr()"""
    )

    float3[] extent = [(-1.0, -1.0, -1.0), (1.0, 1.0, 1.0)] (
        doc = """Extent is re-defined on Cube only to provide a fallback value.
        \\sa UsdGeomGprim::GetExtentAttr()."""
    )

}

class Sphere "Sphere" (
    inherits = </Gprim>
    doc = """Defines a primitive sphere centered at the origin.
    
    The fallback values for Cube, Sphere, Cone, and Cylinder are set so that
    they all pack into the same volume/bounds."""
    customData = {
        dictionary extraPlugInfo = {
            bool implementsComputeExtent = true
        }
    }
) {
    double radius = 1.0 (
        doc = """Indicates the sphere's radius.  If you
        author \\em radius you must also author \\em extent.
        
        \\sa GetExtentAttr()"""
    )

    float3[] extent = [(-1.0, -1.0, -1.0), (1.0, 1.0, 1.0)] (
        doc = """Extent is re-defined on Sphere only to provide a fallback
        value. \\sa UsdGeomGprim::GetExtentAttr()."""
    )
}

class Cylinder "Cylinder" (
    inherits = </Gprim>
    doc = """Defines a primitive cylinder with closed ends, centered at the 
    origin, whose spine is along the specified \\em axis.
    
    The fallback values for Cube, Sphere, Cone, and Cylinder are set so that
    they all pack into the same volume/bounds."""
    customData = {
        dictionary extraPlugInfo = {
            bool implementsComputeExtent = true
        }
    }
) {
    double height = 2 (
        doc = """The size of the cylinder's spine along the specified
        \\em axis.  If you author \\em height you must also author \\em extent.
        
        \\sa GetExtentAttr()"""
    )
    double radius = 1.0 (
        doc = """The radius of the cylinder. If you author \\em radius
        you must also author \\em extent.
        
        \\sa GetExtentAttr()"""
    )
    uniform token axis = "Z" (
        allowedTokens = ["X", "Y", "Z"]
        doc = """The axis along which the spine of the cylinder is aligned"""
    )

    float3[] extent = [(-1.0, -1.0, -1.0), (1.0, 1.0, 1.0)] (
        doc = """Extent is re-defined on Cylinder only to provide a fallback
        value. \\sa UsdGeomGprim::GetExtentAttr()."""
    )
}

class Capsule "Capsule" (
    inherits = </Gprim>
    doc = """Defines a primitive capsule, i.e. a cylinder capped by two half
    spheres, centered at the origin, whose spine is along the specified
    \\em axis."""
    customData = {
        dictionary extraPlugInfo = {
            bool implementsComputeExtent = true
        }
    }
) {
    double height = 1.0 (
        doc = """The size of the capsule's spine along the specified
        \\em axis excluding the size of the two half spheres, i.e.
        the size of the cylinder portion of the capsule.
        If you author \\em height you must also author \\em extent.
        \\sa GetExtentAttr()"""
    )
    double radius = 0.5 (
        doc = """The radius of the capsule.  If you
        author \\em radius you must also author \\em extent.
        
        \\sa GetExtentAttr()"""
    )
    uniform token axis = "Z" (
        allowedTokens = ["X", "Y", "Z"]
        doc = """The axis along which the spine of the capsule is aligned"""
    )

    float3[] extent = [(-0.5, -0.5, -1.0), (0.5, 0.5, 1.0)] (
        doc = """Extent is re-defined on Capsule only to provide a fallback
        value. \\sa UsdGeomGprim::GetExtentAttr()."""
    )
}

class Cone "Cone" (
    inherits = </Gprim>
    doc = """Defines a primitive cone, centered at the origin, whose spine
    is along the specified \\em axis, with the apex of the cone pointing
    in the direction of the positive axis.
    
    The fallback values for Cube, Sphere, Cone, and Cylinder are set so that
    they all pack into the same volume/bounds."""
    customData = {
        dictionary extraPlugInfo = {
            bool implementsComputeExtent = true
        }
    }
) {
    double height = 2.0 (
        doc = """The size of the cone's spine along the specified
        \\em axis.  If you author \\em height you must also author \\em extent.
        
        \\sa GetExtentAttr()"""
    )
    double radius = 1.0 (
        doc = """The radius of the cone.  If you
        author \\em radius you must also author \\em extent.
        
        \\sa GetExtentAttr()"""
    )
    uniform token axis = "Z" (
        allowedTokens = ["X", "Y", "Z"]
        doc = """The axis along which the spine of the cone is aligned"""
    )

    float3[] extent = [(-1.0, -1.0, -1.0), (1.0, 1.0, 1.0)] (
        doc = """Extent is re-defined on Cone only to provide a fallback
        value. \\sa UsdGeomGprim::GetExtentAttr()."""
    )
}

class "PointBased" (
    doc = """Base class for all UsdGeomGprims that possess points,
    providing common attributes such as normals and velocities."""
    
    inherits = </Gprim>
    customData = {
        dictionary extraPlugInfo = {
            bool implementsComputeExtent = true
        }
    }
) {
    # positional
    point3f[] points (
        doc = """The primary geometry attribute for all PointBased
        primitives, describes points in (local) space."""
    )
    
    vector3f[] velocities (
        doc = """If provided, 'velocities' should be used by renderers to 

        compute positions between samples for the 'points' attribute, rather
        than interpolating between neighboring 'points' samples.  This is the
        only reasonable means of computing motion blur for topologically
        varying PointBased primitives.  It follows that the length of each
        'velocities' sample must match the length of the corresponding
        'points' sample.  Velocity is measured in position units per second,
        as per most simulation software. To convert to position units per
        UsdTimeCode, divide by UsdStage::GetTimeCodesPerSecond().
        
        See also \\ref UsdGeom_VelocityInterpolation .""" 
    )

    # shaping
    normal3f[] normals (
        doc = """Provide an object-space orientation for individual points, 
        which, depending on subclass, may define a surface, curve, or free 
        points.  Note that 'normals' should not be authored on any Mesh that
        is subdivided, since the subdivision algorithm will define its own
        normals. 'normals' is not a generic primvar, but the number of elements
        in this attribute will be determined by its 'interpolation'.  See
        \\ref SetNormalsInterpolation() . If 'normals' and 'primvars:normals'
        are both specified, the latter has precedence."""
    )
}

class Mesh "Mesh" (
    inherits = </PointBased>
    customData = {
        string extraIncludes = """
#include "pxr/usd/usd/timeCode.h" """
    }
    doc="""Encodes a mesh surface whose definition and feature-set
    will converge with that of OpenSubdiv, http://graphics.pixar.com/opensubdiv/docs/subdivision_surfaces.html. Current exceptions/divergences include:
    
    1. Certain interpolation ("tag") parameters not yet supported
    
    2. Does not (as of 9/2014) yet support hierarchical edits.  We do intend
    to provide some encoding in a future version of the schema.
    
    A key property of this mesh schema is that it encodes both subdivision
    surfaces, and non-subdived "polygonal meshes", by varying the
    \\em subdivisionScheme attribute.
    
    \\section UsdGeom_Mesh_Normals A Note About Normals

    Normals should not be authored on a subdivided mesh, since subdivision
    algorithms define their own normals. They should only be authored for
    polygonal meshes.

    The 'normals' attribute inherited from UsdGeomPointBased is not a generic
    primvar, but the number of elements in this attribute will be determined by
    its 'interpolation'.  See \\ref UsdGeomPointBased::GetNormalsInterpolation() .
    If 'normals' and 'primvars:normals' are both specified, the latter has
    precedence."""
) {
    #
    # Common Properties
    #
    int[] faceVertexIndices (
        doc = """Flat list of the index (into the 'points' attribute) of each
        vertex of each face in the mesh.  If this attribute has more than
        one timeSample, the mesh is considered to be topologically varying."""
    )
    
    int[] faceVertexCounts (
        doc = """Provides the number of vertices in each face of the mesh, 
        which is also the number of consecutive indices in 'faceVertexIndices'
        that define the face.  The length of this attribute is the number of
        faces in the mesh.  If this attribute has more than
        one timeSample, the mesh is considered to be topologically varying."""
    )

    #
    # Subdiv Properties
    #
    
    uniform token subdivisionScheme = "catmullClark" (
        allowedTokens = ["catmullClark", "loop", "bilinear", "none"]
        doc = """The subdivision scheme to be applied to the surface.  
        Valid values are "catmullClark" (the default), "loop", "bilinear", and
        "none" (i.e. a polymesh with no subdivision - the primary difference
        between schemes "bilinear" and "none" is that bilinearly subdivided
        meshes can be considered watertight, whereas there is no such guarantee
        for un-subdivided polymeshes, and more mesh features (e.g. holes) may
        apply to bilinear meshes but not polymeshes.  Polymeshes \\em may be
        lighterweight and faster to render, depending on renderer and render
        mode.)""")

    token interpolateBoundary = "edgeAndCorner" (
        allowedTokens = ["none", "edgeAndCorner", "edgeOnly"]
        doc = """Specifies how interpolation boundary face edges are
        interpolated. Valid values are "none", 
        "edgeAndCorner" (the default), or "edgeOnly".""")

    token faceVaryingLinearInterpolation = "cornersPlus1" (
        allowedTokens = ["all", "none", "boundaries", "cornersOnly", 
                         "cornersPlus1", "cornersPlus2"]
        doc = """Specifies how face varying data is interpolated.  Valid values
        are "all" (no smoothing), "cornersPlus1" (the default, Smooth UV),
        "none" (Same as "cornersPlus1" but does not infer the presence 
        of corners where two faceVarying edges meet at a single face), or 
        "boundaries" (smooth only near vertices that are not at a
        discontinuous boundary).

        See http://graphics.pixar.com/opensubdiv/docs/subdivision_surfaces.html#face-varying-interpolation-rules""")

    token triangleSubdivisionRule = "catmullClark" (
        allowedTokens = ["catmullClark", "smooth"]
        doc = """Specifies what weights are used during triangle subdivision for
        the Catmull-Clark scheme. Valid values are "catmullClark" (the default) 
        and "smooth".

        See http://graphics.pixar.com/opensubdiv/docs/subdivision_surfaces.html#triangle-subdivision-rule""")
    
    int[] holeIndices = [] (
        doc = """The face indices (indexing into the 'faceVertexCounts'
        attribute) of all faces that should be made invisible.""")

    int[] cornerIndices = [] (
        doc = """The vertex indices of all vertices that are sharp corners.""")

    float[] cornerSharpnesses = [] (
        doc = """The sharpness values for corners: each corner gets a single
        sharpness value (Usd.Mesh.SHARPNESS_INFINITE for a perfectly sharp
        corner), so the size of this array must match that of
        'cornerIndices'""")

    int[] creaseIndices = [] (
        doc = """The indices of all vertices forming creased edges.  The size of 
        this array must be equal to the sum of all elements of the 
        'creaseLengths' attribute.""")

    int[] creaseLengths = [] (
        doc = """The length of this array specifies the number of creases on the
        surface. Each element gives the number of (must be adjacent) vertices in
        each crease, whose indices are linearly laid out in the 'creaseIndices'
        attribute. Since each crease must be at least one edge long, each
        element of this array should be greater than one.""")

    float[] creaseSharpnesses = [] (
        doc = """The per-crease or per-edge sharpness for all creases
        (Usd.Mesh.SHARPNESS_INFINITE for a perfectly sharp crease).  Since
        'creaseLengths' encodes the number of vertices in each crease, the
        number of elements in this array will be either len(creaseLengths) or
        the sum over all X of (creaseLengths[X] - 1). Note that while
        the RI spec allows each crease to have either a single sharpness
        or a value per-edge, USD will encode either a single sharpness
        per crease on a mesh, or sharpnesses for all edges making up
        the creases on a mesh.""")
}

class GeomSubset "GeomSubset" (
    inherits = </Typed>
    doc = """Encodes a subset of a piece of geometry (i.e. a UsdGeomImageable) 
    as a set of indices. Currently only supports encoding of face-subsets, but 
    could be extended in the future to support subsets representing edges, 
    segments, points etc.

    To apply to a geometric prim, a GeomSubset prim must be defined as a 
    child of it in namespace. This restriction makes it easy and efficient 
    to discover subsets of a prim. We might want to relax this restriction if 
    it's common to have multiple <b>families</b> of subsets on a gprim and if 
    it's useful to be able to organize subsets belonging to a </b>family</b> 
    under a common scope. See 'familyName' attribute for more info on defining 
    a family of subsets.

    Note that a GeomSubset isn't an imageable (i.e. doesn't derive from
    UsdGeomImageable). So, you can't author <b>visibility</b> for it or 
    override its <b>purpose</b>.

    Materials are bound to GeomSubsets just as they are for regular 
    geometry using API available in UsdShade (UsdShadeMaterial::Bind).
"""
    customData = {
        string className = "Subset"
        string extraIncludes = """
#include "pxr/base/tf/token.h"
#include "pxr/usd/usdGeom/imageable.h"
"""
    }
)
{
    uniform token elementType = "face" (
        allowedTokens = ["face"]
        doc = """The type of element that the indices target. Currently only 
        allows "face" and defaults to it."""
    )
    int[] indices = [] (
        doc = """The set of indices included in this subset. The indices need not 
        be sorted, but the same index should not appear more than once."""
    )
    uniform token familyName = "" (
        doc = """The name of the family of subsets that this subset belongs to. 
        This is optional and is primarily useful when there are multiple 
        families of subsets under a geometric prim. In some cases, this could 
        also be used for achieving proper roundtripping of subset data between 
        DCC apps.
        When multiple subsets belonging to a prim have the same familyName, they 
        are said to belong to the family. A <i>familyType</i> value can be 
        encoded on the owner of a family of subsets as a token using the static 
        method UsdGeomSubset::SetFamilyType(). "familyType" can have one of the 
        following values:
        <ul><li><b>UsdGeomTokens->partition</b>: implies that every element of 
        the whole geometry appears exactly once in only one of the subsets
        belonging to the family.</li>
        <li><b>UsdGeomTokens->nonOverlapping</b>: an element that appears in one 
        subset may not appear in any other subset belonging to the family.</li>
        <li><b>UsdGeomTokens->unrestricted</b>: implies that there are no
        restrictions w.r.t. the membership of elements in the subsets. They 
        could be overlapping and the union of all subsets in the family may 
        not represent the whole.</li>
        </ul>
        \\note The validity of subset data is not enforced by the authoring 
        APIs, however they can be checked using UsdGeomSubset::ValidateFamily().
        """
    )
}

class NurbsPatch "NurbsPatch" (
    inherits = </PointBased>
    doc = """Encodes a rational or polynomial non-uniform B-spline
    surface, with optional trim curves.
    
    The encoding mostly follows that of RiNuPatch and RiTrimCurve: 
    https://renderman.pixar.com/resources/current/RenderMan/geometricPrimitives.html#rinupatch , with some minor renaming and coalescing for clarity.
    
    The layout of control vertices in the \\em points attribute inherited
    from UsdGeomPointBased is row-major with U considered rows, and V columns.
    
    \\anchor UsdGeom_NurbsPatch_Form
    <b>NurbsPatch Form</b>
    
    The authored points, orders, knots, weights, and ranges are all that is
    required to render the nurbs patch.  However, the only way to model closed
    surfaces with nurbs is to ensure that the first and last control points
    along the given axis are coincident.  Similarly, to ensure the surface is
    not only closed but also C2 continuous, the last \\em order - 1 control
    points must be (correspondingly) coincident with the first \\em order - 1
    control points, and also the spacing of the last corresponding knots
    must be the same as the first corresponding knots.
    
    <b>Form</b> is provided as an aid to interchange between modeling and
    animation applications so that they can robustly identify the intent with
    which the surface was modelled, and take measures (if they are able) to
    preserve the continuity/concidence constraints as the surface may be rigged
    or deformed.  
    \\li An \\em open-form NurbsPatch has no continuity constraints.
    \\li A \\em closed-form NurbsPatch expects the first and last control points
    to overlap
    \\li A \\em periodic-form NurbsPatch expects the first and last
    \\em order - 1 control points to overlap.
    
    <b>Nurbs vs Subdivision Surfaces</b>
    
    Nurbs are an important modeling primitive in CAD/CAM tools and early
    computer graphics DCC's.  Because they have a natural UV parameterization
    they easily support "trim curves", which allow smooth shapes to be
    carved out of the surface.
    
    However, the topology of the patch is always rectangular, and joining two 
    nurbs patches together (especially when they have differing numbers of
    spans) is difficult to do smoothly.  Also, nurbs are not supported by
    the Ptex texturing technology (http://ptex.us).
    
    Neither of these limitations are shared by subdivision surfaces; therefore,
    although they do not subscribe to trim-curve-based shaping, subdivs are
    often considered a more flexible modeling primitive.
    """
) {
    int uVertexCount (
        doc = """Number of vertices in the U direction.  Should be at least as
        large as uOrder."""
    )

    int vVertexCount (
        doc = """Number of vertices in the V direction.  Should be at least as
        large as vOrder."""
    )

    int uOrder (
        doc = """Order in the U direction.  Order must be positive and is
        equal to the degree of the polynomial basis to be evaluated, plus 1."""
    )

    int vOrder (
        doc = """Order in the V direction.  Order must be positive and is
        equal to the degree of the polynomial basis to be evaluated, plus 1."""
    )

    double[] uKnots (
        doc = """Knot vector for U direction providing U parameterization.
        The length of this array must be ( uVertexCount + uOrder ), and its
        entries must take on monotonically increasing values."""  
    )

    double[] vKnots (
        doc = """Knot vector for V direction providing U parameterization.
        The length of this array must be ( vVertexCount + vOrder ), and its
        entries must take on monotonically increasing values."""  
    )

    uniform token uForm = "open" (
        allowedTokens = ["open", "closed", "periodic"]
        doc = """Interpret the control grid and knot vectors as representing
        an open, geometrically closed, or geometrically closed and C2 continuous
        surface along the U dimension.
        \\sa \\ref UsdGeom_NurbsPatch_Form "NurbsPatch Form" """
    )
        
    uniform token vForm = "open" (
        allowedTokens = ["open", "closed", "periodic"]
        doc = """Interpret the control grid and knot vectors as representing
        an open, geometrically closed, or geometrically closed and C2 continuous
        surface along the V dimension.
        \\sa \\ref UsdGeom_NurbsPatch_Form "NurbsPatch Form" """
    )
        
    # Alembic's NuPatch does not encode these... wonder how they 
    # get away with that?  Just assume it's the full range, presumably.
    double2 uRange (
        doc = """Provides the minimum and maximum parametric values (as defined
        by uKnots) over which the surface is actually defined.  The minimum
        must be less than the maximum, and greater than or equal to the
        value of uKnots[uOrder-1].  The maxium must be less than or equal
        to the last element's value in uKnots."""
    )

    double2 vRange (
        doc = """Provides the minimum and maximum parametric values (as defined
        by vKnots) over which the surface is actually defined.  The minimum
        must be less than the maximum, and greater than or equal to the
        value of vKnots[vOrder-1].  The maxium must be less than or equal
        to the last element's value in vKnots."""
    )

    double[] pointWeights (
        doc = """Optionally provides "w" components for each control point,
        thus must be the same length as the points attribute.  If authored,
        the patch will be rational.  If unauthored, the patch will be
        polynomial, i.e. weight for all points is 1.0.
        \\note Some DCC's pre-weight the \\em points, but in this schema, 
        \\em points are not pre-weighted."""
    )

    int[] trimCurve:counts (
        doc = """Each element specifies how many curves are present in each
        "loop" of the trimCurve, and the length of the array determines how
        many loops the trimCurve contains.  The sum of all elements is the
        total nuber of curves in the trim, to which we will refer as 
        \\em nCurves in describing the other trim attributes."""
    )
    
    int[] trimCurve:orders (
        doc = """Flat list of orders for each of the \\em nCurves curves."""
    )
    
    int[] trimCurve:vertexCounts (
        doc = """Flat list of number of vertices for each of the
         \\em nCurves curves."""
    )

    double[] trimCurve:knots (
        doc = """Flat list of parametric values for each of the
        \\em nCurves curves.  There will be as many knots as the sum over
        all elements of \\em vertexCounts plus the sum over all elements of
        \\em orders."""
    )

    double2[] trimCurve:ranges (
        doc = """Flat list of minimum and maximum parametric values 
        (as defined by \\em knots) for each of the \\em nCurves curves."""
    )

    double3[] trimCurve:points (
        doc = """Flat list of homogeneous 2D points (u, v, w) that comprise
        the \\em nCurves curves.  The number of points should be equal to the
        um over all elements of \\em vertexCounts."""
    )

}

class "Curves" (
    customData = {
        dictionary extraPlugInfo = {
            bool implementsComputeExtent = true
        }
    }
    inherits = </PointBased>
    doc = """Base class for BasisCurves and NurbsCurves.  The BasisCurves
             schema is designed to be analagous to offline renderers' notion
             of batched curves (such as the classical RIB definition via
             Basis and Curves statements), while the NurbsCurve schema
             is designed to be analgous to the NURBS curves found in packages
             like Maya and Houdini while retaining their consistency with the 
             RenderMan specification for NURBS Patches."""
) {
    # topology attributes
    int[] curveVertexCounts (
        doc = """Curves-derived primitives can represent multiple distinct,
        potentially disconnected curves.  The length of 'curveVertexCounts'
        gives the number of such curves, and each element describes the
        number of vertices in the corresponding curve"""
    )
    
    # shaping attributes
    float[] widths (
        doc = """Provides width specification for the curves, whose application
        will depend on whether the curve is oriented (normals are defined for
        it), in which case widths are "ribbon width", or unoriented, in which
        case widths are cylinder width.  'widths' is not a generic Primvar,
        but the number of elements in this attribute will be determined by
        its 'interpolation'.  See \\ref SetWidthsInterpolation() .  If 'widths'
        and 'primvars:widths' are both specified, the latter has precedence."""
    )
}

class BasisCurves "BasisCurves" (
    inherits = </Curves>
    doc = """BasisCurves are a batched curve representation analogous to the
    classic RIB definition via Basis and Curves statements. BasisCurves are
    often used to render dense aggregate geometry like hair or grass.

    A 'matrix' and 'vstep' associated with the \\em basis are used to
    interpolate the vertices of a cubic BasisCurves. (The basis attribute
    is unused for linear BasisCurves.)
    
    A single prim may have many curves whose count is determined implicitly by
    the length of the \\em curveVertexCounts vector.  Each individual curve is
    composed of one or more segments. Each segment is defined by four vertices
    for cubic curves and two vertices for linear curves. See the next section
    for more information on how to map curve vertex counts to segment counts.

    \\section UsdGeomBasisCurves_Segment Segment Indexing
    Interpolating a curve requires knowing how to decompose it into its 
    individual segments.

    The segments of a cubic curve are determined by the vertex count,
    the \\em wrap (periodicity), and the vstep of the basis. For linear
    curves, the basis token is ignored and only the vertex count and
    wrap are needed.
    
    cubic basis   | vstep
    ------------- | ------
    bezier        | 3
    catmullRom    | 1
    bspline       | 1
    
    The first segment of a cubic (nonperiodic) curve is always defined by its
    first four points. The vstep is the increment used to determine what
    vertex indices define the next segment.  For a two segment (nonperiodic)
    bspline basis curve (vstep = 1), the first segment will be defined by
    interpolating vertices [0, 1, 2, 3] and the second segment will be defined
    by [1, 2, 3, 4].  For a two segment bezier basis curve (vstep = 3), the
    first segment will be defined by interpolating vertices [0, 1, 2, 3] and
    the second segment will be defined by [3, 4, 5, 6].  If the vstep is not
    one, then you must take special care to make sure that the number of cvs
    properly divides by your vstep. (The indices described are relative to
    the initial vertex index for a batched curve.)

    For periodic curves, at least one of the curve's initial vertices are
    repeated to close the curve. For cubic curves, the number of vertices
    repeated is '4 - vstep'. For linear curves, only one vertex is repeated
    to close the loop.
    
    Pinned curves are a special case of nonperiodic curves that only affects
    the behavior of cubic Bspline and Catmull-Rom curves. To evaluate or render
    pinned curves, a client must effectively add 'phantom points' at the 
    beginning and end of every curve in a batch.  These phantom points
    are injected to ensure that the inteprolated curve begins at P[0] and
    ends at P[n-1].
    
    For a curve with initial point P[0] and last point P[n-1], the phantom
    points are defined as.
    P[-1]  = 2 * P[0] - P[1]
    P[n] = 2 * P[n-1] - P[n-2]

    Pinned cubic curves will (usually) have to be unpacked into the standard
    nonperiodic representation before rendering. This unpacking can add some 
    additional overhead. However, using pinned curves reduces the amount of
    data recorded in a scene and (more importantly) better records the
    authors' intent for interchange.

    \\note The additional phantom points mean that the minimum curve vertex
    count for cubic bspline and catmullRom curves is 2.

    Linear curve segments are defined by two vertices.
    A two segment linear curve's first segment would be defined by
    interpolating vertices [0, 1]. The second segment would be defined by 
    vertices [1, 2]. (Again, for a batched curve, indices are relative to
    the initial vertex index.)

    When validating curve topology, each renderable entry in the
    curveVertexCounts vector must pass this check.

    type    | wrap                        | validitity
    ------- | --------------------------- | ----------------
    linear  | nonperiodic                 | curveVertexCounts[i] > 2
    linear  | periodic                    | curveVertexCounts[i] > 3
    cubic   | nonperiodic                 | (curveVertexCounts[i] - 4) % vstep == 0
    cubic   | periodic                    | (curveVertexCounts[i]) % vstep == 0
    cubic   | pinned (catmullRom/bspline) | (curveVertexCounts[i] - 2) >= 0

    \\section UsdGeomBasisCurves_BasisMatrix Cubic Vertex Interpolation
 
    \\image html USDCurveBasisMatrix.png width=750
        
    \\section UsdGeomBasisCurves_Linear Linear Vertex Interpolation
 
    Linear interpolation is always used on curves of type linear.
    't' with domain [0, 1], the curve is defined by the equation 
    P0 * (1-t) + P1 * t. t at 0 describes the first point and t at 1 describes
    the end point.

    \\section UsdGeomBasisCurves_PrimvarInterpolation Primvar Interpolation
        
    For cubic curves, primvar data can be either interpolated cubically between 
    vertices or linearly across segments.  The corresponding token
    for cubic interpolation is 'vertex' and for linear interpolation is
    'varying'.  Per vertex data should be the same size as the number
    of vertices in your curve.  Segment varying data is dependent on the 
    wrap (periodicity) and number of segments in your curve.  For linear curves,
    varying and vertex data would be interpolated the same way.  By convention 
    varying is the preferred interpolation because of the association of 
    varying with linear interpolation.
    
    \\image html USDCurvePrimvars.png 

    To convert an entry in the curveVertexCounts vector into a segment count 
    for an individual curve, apply these rules.  Sum up all the results in
    order to compute how many total segments all curves have.

    The following tables describe the expected segment count for the 'i'th
    curve in a curve batch as well as the entire batch. Python syntax
    like '[:]' (to describe all members of an array) and 'len(...)' 
    (to describe the length of an array) are used.

    type    | wrap                        | curve segment count                    | batch segment count                                                       
    ------- | --------------------------- | -------------------------------------- | --------------------------
    linear  | nonperiodic                 | curveVertexCounts[i] - 1               | sum(curveVertexCounts[:]) - len(curveVertexCounts)
    linear  | periodic                    | curveVertexCounts[i]                   | sum(curveVertexCounts[:])
    cubic   | nonperiodic                 | (curveVertexCounts[i] - 4) / vstep + 1 | sum(curveVertexCounts[:] - 4) / vstep + len(curveVertexCounts)
    cubic   | periodic                    | curveVertexCounts[i] / vstep           | sum(curveVertexCounts[:]) / vstep
    cubic   | pinned (catmullRom/bspline) | (curveVertexCounts[i] - 2) + 1         | sum(curveVertexCounts[:] - 2) + len(curveVertexCounts)
 
    The following table descrives the expected size of varying
    (linearly interpolated) data, derived from the segment counts computed
    above.

    wrap                | curve varying count          | batch varying count
    ------------------- | ---------------------------- | ------------------------------------------------
    nonperiodic/pinned  | segmentCounts[i] + 1         | sum(segmentCounts[:]) + len(curveVertexCounts)
    periodic            | segmentCounts[i]             | sum(segmentCounts[:])

    Both curve types additionally define 'constant' interpolation for the
    entire prim and 'uniform' interpolation as per curve data.

 
    \\note Take care when providing support for linearly interpolated data for
    cubic curves. Its shape doesn't provide a one to one mapping with either
    the number of curves (like 'uniform') or the number of vertices (like
    'vertex') and so it is often overlooked. This is the only primitive in
    UsdGeom (as of this writing) where this is true. For meshes, while they
    use different interpolation methods, 'varying' and 'vertex' are both
    specified per point. It's common to assume that curves follow a similar
    pattern and build in structures and language for per primitive, per
    element, and per point data only to come upon these arrays that don't 
    quite fit into either of those categories. It is
    also common to conflate 'varying' with being per segment data and use the
    segmentCount rules table instead of its neighboring varying data table
    rules. We suspect that this is because for the common case of
    nonperiodic cubic curves, both the provided segment count and varying data
    size formula end with '+ 1'. While debugging, users may look at the double
    '+ 1' as a mistake and try to remove it.  We take this time to enumerate
    these issues because we've fallen into them before and hope that we save
    others time in their own implementations.

    As an example of deriving per curve segment and varying primvar data counts from
    the wrap, type, basis, and curveVertexCount, the following table is provided.

    wrap          | type    | basis   | curveVertexCount  | curveSegmentCount  | varyingDataCount
    ------------- | ------- | ------- | ----------------- | ------------------ | -------------------------
    nonperiodic   | linear  | N/A     | [2 3 2 5]         | [1 2 1 4]          | [2 3 2 5]
    nonperiodic   | cubic   | bezier  | [4 7 10 4 7]      | [1 2 3 1 2]        | [2 3 4 2 3]
    nonperiodic   | cubic   | bspline | [5 4 6 7]         | [2 1 3 4]          | [3 2 4 5]
    periodic      | cubic   | bezier  | [6 9 6]           | [2 3 2]            | [2 3 2]
    periodic      | linear  | N/A     | [3 7]             | [3 7]              | [3 7]

    \\section UsdGeomBasisCurves_TubesAndRibbons Tubes and Ribbons
    
    The strictest definition of a curve as an infinitely thin wire is not 
    particularly useful for describing production scenes. The additional
    \\em widths and \\em normals attributes can be used to describe cylindrical
    tubes and or flat oriented ribbons.

    Curves with only widths defined are imaged as tubes with radius
    'width / 2'. Curves with both widths and normals are imaged as ribbons
    oriented in the direction of the interpolated normal vectors.

    While not technically UsdGeomPrimvars, widths and normals
    also have interpolation metadata. It's common for authored widths to have
    constant, varying, or vertex interpolation 
    (see UsdGeomCurves::GetWidthsInterpolation()).  It's common for
    authored normals to have varying interpolation 
    (see UsdGeomPointBased::GetNormalsInterpolation()).

    \\image html USDCurveHydra.png

    The file used to generate these curves can be found in
    pxr/extras/examples/usdGeomExamples/basisCurves.usda.  It's provided
    as a reference on how to properly image both tubes and ribbons. The first
    row of curves are linear; the second are cubic bezier. (We aim in future
    releases of HdSt to fix the discontinuity seen with broken tangents to
    better match offline renderers like RenderMan.) The yellow and violet
    cubic curves represent cubic vertex width interpolation for which there is
    no equivalent for linear curves.
    
    \\note How did this prim type get its name?  This prim is a portmanteau of
    two different statements in the original RenderMan specification:
    'Basis' and 'Curves'.
"""
) {
    # interpolation attributes
    uniform token type  = "cubic" (
        allowedTokens = ["linear", "cubic"]
        doc = """Linear curves interpolate linearly between two vertices.  
        Cubic curves use a basis matrix with four vertices to interpolate a segment.""")

    uniform token basis = "bezier" (
        allowedTokens = ["bezier", "bspline", "catmullRom", "hermite", "power"]
        doc = """The basis specifies the vstep and matrix used for cubic interpolation.""")

    uniform token wrap = "nonperiodic" (
        allowedTokens = ["nonperiodic", "periodic", "pinned"]
        doc = """If wrap is set to periodic, the curve when rendered will 
        repeat the initial vertices (dependent on the vstep) to close the
        curve. If wrap is set to 'pinned', phantom points may be created
        to ensure that the curve interpolation starts at P[0] and ends at P[n-1].
        """)
}

class NurbsCurves "NurbsCurves" (
    inherits = </Curves>
    doc = """This schema is analagous to NURBS Curves in packages like Maya
    and Houdini, often used for interchange of rigging and modeling curves.  
    Unlike Maya, this curve spec supports batching of multiple curves into a 
    single prim, widths, and normals in the schema.  Additionally, we require 
    'numSegments + 2 * degree + 1' knots (2 more than maya does).  This is to
    be more consistent with RenderMan's NURBS patch specification.  
    
    To express a periodic curve:
    - knot[0] = knot[1] - (knots[-2] - knots[-3]; 
    - knot[-1] = knot[-2] + (knot[2] - knots[1]);
    
    To express a nonperiodic curve:
    - knot[0] = knot[1];
    - knot[-1] = knot[-2];
    
    In spite of these slight differences in the spec, curves generated in Maya
    should be preserved when roundtripping.
    
    \\em order and \\em range, when representing a batched NurbsCurve should be
    authored one value per curve.  \\em knots should be the concatentation of
    all batched curves."""
) {
    # topology attributes
    int[] order = [] (
        doc = """Order of the curve.  Order must be positive and is
        equal to the degree of the polynomial basis to be evaluated, plus 1.
        Its value for the 'i'th curve must be less than or equal to
        curveVertexCount[i]""")

    # interpolation attributes
    double[] knots (
        doc = """Knot vector providing curve parameterization.
        The length of the slice of the array for the ith curve 
        must be ( curveVertexCount[i] + order[i] ), and its
        entries must take on monotonically increasing values.""")
        
    double2[] ranges (
        doc = """Provides the minimum and maximum parametric values (as defined
        by knots) over which the curve is actually defined.  The minimum must 
        be less than the maximum, and greater than or equal to the value of the 
        knots['i'th curve slice][order[i]-1]. The maxium must be less 
        than or equal to the last element's value in knots['i'th curve slice].
	Range maps to (vmin, vmax) in the RenderMan spec.""")
}

class Points "Points" (
    customData = {
        dictionary extraPlugInfo = {
            bool implementsComputeExtent = true
        }
    }
    inherits = </PointBased>
    doc = """Points are analogous to the <A HREF="https://renderman.pixar.com/resources/current/RenderMan/appnote.18.html">RiPoints spec</A>.  
    
    Points can be an efficient means of storing and rendering particle
    effects comprised of thousands or millions of small particles.  Points
    generally receive a single shading sample each, which should take 
    \\em normals into account, if present.

    While not technically UsdGeomPrimvars, the widths and normals also
    have interpolation metadata.  It's common for authored widths and normals
    to have constant or varying interpolation."""

) {
    # shaping attributes
    float[] widths (
        doc = """Widths are defined as the \\em diameter of the points, in 
                 object space.  'widths' is not a generic Primvar, but
                 the number of elements in this attribute will be determined by
                 its 'interpolation'.  See \\ref SetWidthsInterpolation() .  If
                 'widths' and 'primvars:widths' are both specified, the latter
                 has precedence."""
    )
    
    int64[] ids (
        doc = """Ids are optional; if authored, the ids array should be the same
                 length as the points array, specifying (at each timesample if
                 point identities are changing) the id of each point. The
                 type is signed intentionally, so that clients can encode some
                 binary state on Id'd points without adding a separate 
                 primvar."""
    )
}

class PointInstancer "PointInstancer" (
    doc = """Encodes vectorized instancing of multiple, potentially
    animated, prototypes (object/instance masters), which can be arbitrary
    prims/subtrees on a UsdStage.
    
    PointInstancer is a "multi instancer", as it allows multiple prototypes
    to be scattered among its "points".  We use a UsdRelationship
    \\em prototypes to identify and order all of the possible prototypes, by
    targeting the root prim of each prototype.  The ordering imparted by
    relationships associates a zero-based integer with each prototype, and
    it is these integers we use to identify the prototype of each instance,
    compactly, and allowing prototypes to be swapped out without needing to
    reauthor all of the per-instance data.
    
    The PointInstancer schema is designed to scale to billions of instances,
    which motivates the choice to split the per-instance transformation into
    position, (quaternion) orientation, and scales, rather than a
    4x4 matrix per-instance.  In addition to requiring fewer bytes even if
    all elements are authored (32 bytes vs 64 for a single-precision 4x4
    matrix), we can also be selective about which attributes need to animate
    over time, for substantial data reduction in many cases.
    
    Note that PointInstancer is \\em not a Gprim, since it is not a graphical
    primitive by any stretch of the imagination. It \\em is, however,
    Boundable, since we will sometimes want to treat the entire PointInstancer
    similarly to a procedural, from the perspective of inclusion or framing.

    \\section UsdGeomPointInstancer_varyingTopo Varying Instance Identity over Time
    
    PointInstancers originating from simulations often have the characteristic
    that points/instances are "born", move around for some time period, and then
    die (or leave the area of interest). In such cases, billions of instances
    may be birthed over time, while at any \\em specific time, only a much
    smaller number are actually alive.  To encode this situation efficiently,
    the simulator may re-use indices in the instance arrays, when a particle
    dies, its index will be taken over by a new particle that may be birthed in
    a much different location.  This presents challenges both for 
    identity-tracking, and for motion-blur.
    
    We facilitate identity tracking by providing an optional, animatable
    \\em ids attribute, that specifies the 64 bit integer ID of the particle
    at each index, at each point in time.  If the simulator keeps monotonically
    increasing a particle-count each time a new particle is birthed, it will
    serve perfectly as particle \\em ids.
    
    We facilitate motion blur for varying-topology particle streams by
    optionally allowing per-instance \\em velocities and \\em angularVelocities
    to be authored.  If instance transforms are requested at a time between
    samples and either of the velocity attributes is authored, then we will
    not attempt to interpolate samples of \\em positions or \\em orientations.
    If not authored, and the bracketing samples have the same length, then we
    will interpolate.

    \\section UsdGeomPointInstancer_transform Computing an Instance Transform
    
    Each instance's transformation is a combination of the SRT affine transform
    described by its scale, orientation, and position, applied \\em after
    (i.e. less locally) than the transformation computed at the root of the
    prototype it is instancing.  In other words, to put an instance of a 
    PointInstancer into the space of the PointInstancer's parent prim:
    
    1. Apply (most locally) the authored transformation for 
    <em>prototypes[protoIndices[i]]</em>
    2. If *scales* is authored, next apply the scaling matrix from *scales[i]*
    3. If *orientations* is authored: **if *angularVelocities* is authored**, 
    first multiply *orientations[i]* by the unit quaternion derived by scaling 
    *angularVelocities[i]* by the \\ref UsdGeom_PITimeScaling "time differential" 
    from the left-bracketing timeSample for *orientation* to the requested 
    evaluation time *t*, storing the result in *R*, **else** assign *R* 
    directly from *orientations[i]*.  Apply the rotation matrix derived 
    from *R*.
    4. Apply the translation derived from *positions[i]*. If *velocities* is 
    authored, apply the translation deriving from *velocities[i]* scaled by 
    the time differential from the left-bracketing timeSample for *positions* 
    to the requested evaluation time *t*.
    5. Least locally, apply the transformation authored on the PointInstancer 
    prim itself (or the UsdGeomImageable::ComputeLocalToWorldTransform() of the 
    PointInstancer to put the instance directly into world space)

    If neither *velocities* nor *angularVelocities* are authored, we fallback to
    standard position and orientation computation logic (using linear
    interpolation between timeSamples) as described by
    \\ref UsdGeom_VelocityInterpolation .

    \\anchor UsdGeom_PITimeScaling
    <b>Scaling Velocities for Interpolation</b>
    
    When computing time-differentials by which to apply velocity or
    angularVelocity to positions or orientations, we must scale by
    ( 1.0 / UsdStage::GetTimeCodesPerSecond() ), because velocities are recorded
    in units/second, while we are interpolating in UsdTimeCode ordinates.
    
    Additionally, if *motion:velocityScale* is authored or inherited (see
    UsdGeomMotionAPI::ComputeVelocityScale()), it is used to scale both the
    velocity and angular velocity by a constant value during computation. The
    *motion:velocityScale* attribute is encoded by UsdGeomMotionAPI.

    We provide both high and low-level API's for dealing with the
    transformation as a matrix, both will compute the instance matrices using
    multiple threads; the low-level API allows the client to cache unvarying
    inputs so that they need not be read duplicately when computing over
    time.

    See also \\ref UsdGeom_VelocityInterpolation .
    
    \\section UsdGeomPointInstancer_primvars Primvars on PointInstancer
    
    \\ref UsdGeomPrimvar "Primvars" authored on a PointInstancer prim should
    always be applied to each instance with \\em constant interpolation at
    the root of the instance.  When you are authoring primvars on a 
    PointInstancer, think about it as if you were authoring them on a 
    point-cloud (e.g. a UsdGeomPoints gprim).  The same 
    <A HREF="http://renderman.pixar.com/resources/current/rps/appnote.22.html#classSpecifiers">interpolation rules for points</A> apply here, substituting
    "instance" for "point".
    
    In other words, the (constant) value extracted for each instance
    from the authored primvar value depends on the authored \\em interpolation
    and \\em elementSize of the primvar, as follows:
    \\li <b>constant</b> or <b>uniform</b> : the entire authored value of the
    primvar should be applied exactly to each instance.
    \\li <b>varying</b>, <b>vertex</b>, or <b>faceVarying</b>: the first
    \\em elementSize elements of the authored primvar array should be assigned to
    instance zero, the second \\em elementSize elements should be assigned to
    instance one, and so forth.

    
    \\section UsdGeomPointInstancer_masking Masking Instances: "Deactivating" and Invising

    Often a PointInstancer is created "upstream" in a graphics pipeline, and
    the needs of "downstream" clients necessitate eliminating some of the 
    instances from further consideration.  Accomplishing this pruning by 
    re-authoring all of the per-instance attributes is not very attractive,
    since it may mean destructively editing a large quantity of data.  We
    therefore provide means of "masking" instances by ID, such that the 
    instance data is unmolested, but per-instance transform and primvar data
    can be retrieved with the no-longer-desired instances eliminated from the
    (smaller) arrays.  PointInstancer allows two independent means of masking
    instances by ID, each with different features that meet the needs of
    various clients in a pipeline.  Both pruning features' lists of ID's are
    combined to produce the mask returned by ComputeMaskAtTime().
    
    \\note If a PointInstancer has no authored \\em ids attribute, the masking
    features will still be available, with the integers specifying element
    position in the \\em protoIndices array rather than ID.

    \\subsection UsdGeomPointInstancer_inactiveIds InactiveIds: List-edited, Unvarying Masking

    The first masking feature encodes a list of IDs in a list-editable metadatum
    called \\em inactiveIds, which, although it does not have any similar 
    impact to stage population as \\ref UsdPrim::SetActive() "prim activation",
    it shares with that feature that its application is uniform over all time.
    Because it is list-editable, we can \\em sparsely add and remove instances
    from it in many layers.
    
    This sparse application pattern makes \\em inactiveIds a good choice when
    further downstream clients may need to reverse masking decisions made
    upstream, in a manner that is robust to many kinds of future changes to
    the upstream data.
    
    See ActivateId(), ActivateIds(), DeactivateId(), DeactivateIds(), 
    ActivateAllIds()

    \\subsection UsdGeomPointInstancer_invisibleIds invisibleIds: Animatable Masking

    The second masking feature encodes a list of IDs in a time-varying
    Int64Array-valued UsdAttribute called \\em invisibleIds , since it shares
    with \\ref UsdGeomImageable::GetVisibilityAttr() "Imageable visibility"
    the ability to animate object visibility.
    
    Unlike \\em inactiveIds, overriding a set of opinions for \\em invisibleIds
    is not at all straightforward, because one will, in general need to
    reauthor (in the overriding layer) **all** timeSamples for the attribute
    just to change one Id's visibility state, so it cannot be authored
    sparsely.  But it can be a very useful tool for situations like encoding
    pre-computed camera-frustum culling of geometry when either or both of
    the instances or the camera is animated.
    
    See VisId(), VisIds(), InvisId(), InvisIds(), VisAllIds()
     
    \\section UsdGeomPointInstancer_protoProcessing Processing and Not Processing Prototypes
    
    Any prim in the scenegraph can be targetted as a prototype by the
    \\em prototypes relationship.  We do not, however, provide a specific
    mechanism for identifying prototypes as geometry that should not be drawn
    (or processed) in their own, local spaces in the scenegraph.  We
    encourage organizing all prototypes as children of the PointInstancer
    prim that consumes them, and pruning "raw" processing and drawing
    traversals when they encounter a PointInstancer prim; this is what the
    UsdGeomBBoxCache and UsdImaging engines do.
    
    There \\em is a pattern one can deploy for organizing the prototypes
    such that they will automatically be skipped by basic UsdPrim::GetChildren()
    or UsdPrimRange traversals.  Usd prims each have a 
    \\ref Usd_PrimSpecifiers "specifier" of "def", "over", or "class".  The
    default traversals skip over prims that are "pure overs" or classes.  So
    to protect prototypes from all generic traversals and processing, place
    them under a prim that is just an "over".  For example,
    \\code
    01 def PointInstancer "Crowd_Mid"
    02 {
    03     rel prototypes = [ </Crowd_Mid/Prototypes/MaleThin_Business>, </Crowd_Mid/Prototypes/MaleTine_Casual> ]
    04     
    05     over "Prototypes" 
    06     {
    07          def "MaleThin_Business" (
    08              references = [@MaleGroupA/usd/MaleGroupA.usd@</MaleGroupA>]
    09              variants = {
    10                  string modelingVariant = "Thin"
    11                  string costumeVariant = "BusinessAttire"
    12              }
    13          )
    14          { ... }
    15          
    16          def "MaleThin_Casual"
    17          ...
    18     }
    19 }
    \\endcode
    """

    inherits = </Boundable>
    customData = {
        dictionary extraPlugInfo = {
            bool implementsComputeExtent = true
        }
    }
) {
  rel prototypes (
      doc = """<b>Required property</b>. Orders and targets the prototype root 
      prims, which can be located anywhere in the scenegraph that is convenient,
      although we promote organizing prototypes as children of the 
      PointInstancer.  The position of a prototype in this relationship defines
      the value an instance would specify in the \\em protoIndices attribute to 
      instance that prototype. Since relationships are uniform, this property
      cannot be animated."""
  ) 

  int[] protoIndices (
      doc = """<b>Required property</b>. Per-instance index into 
      \\em prototypes relationship that identifies what geometry should be 
      drawn for each instance.  <b>Topology attribute</b> - can be animated, 
      but at a potential performance impact for streaming."""
  )

  int64[] ids (
      doc = """Ids are optional; if authored, the ids array should be the same
      length as the \\em protoIndices array, specifying (at each timeSample if
      instance identities are changing) the id of each instance. The
      type is signed intentionally, so that clients can encode some
      binary state on Id'd instances without adding a separate primvar.
      See also \\ref UsdGeomPointInstancer_varyingTopo"""
  )

  point3f[] positions (
      doc = """<b>Required property</b>. Per-instance position.  See also 
      \\ref UsdGeomPointInstancer_transform ."""
  )

  quath[] orientations (
      doc="""If authored, per-instance orientation of each instance about its 
      prototype's origin, represented as a unit length quaternion, which
      allows us to encode it with sufficient precision in a compact GfQuath.
      
      It is client's responsibility to ensure that authored quaternions are
      unit length; the convenience API below for authoring orientations from
      rotation matrices will ensure that quaternions are unit length, though
      it will not make any attempt to select the "better (for interpolation
      with respect to neighboring samples)" of the two possible quaternions
      that encode the rotation. 
      
      See also \\ref UsdGeomPointInstancer_transform ."""  )

  float3[] scales (
      doc="""If authored, per-instance scale to be applied to 
      each instance, before any rotation is applied.
      
      See also \\ref UsdGeomPointInstancer_transform ."""
  )

  vector3f[] velocities (
       doc = """If provided, per-instance 'velocities' will be used to 
       compute positions between samples for the 'positions' attribute,
       rather than interpolating between neighboring 'positions' samples.
       Velocities should be considered mandatory if both \\em protoIndices
       and \\em positions are animated.  Velocity is measured in position
       units per second, as per most simulation software. To convert to
       position units per UsdTimeCode, divide by
       UsdStage::GetTimeCodesPerSecond().

       See also \\ref UsdGeomPointInstancer_transform, 
       \\ref UsdGeom_VelocityInterpolation ."""
  )

  vector3f[] angularVelocities (
      doc="""If authored, per-instance angular velocity vector to be used for
      interoplating orientations.  Angular velocities should be considered
      mandatory if both \\em protoIndices and \\em orientations are animated.
      Angular velocity is measured in <b>degrees</b> per second. To convert
      to degrees per UsdTimeCode, divide by
      UsdStage::GetTimeCodesPerSecond().
      
      See also \\ref UsdGeomPointInstancer_transform ."""
  )
  
  int64[] invisibleIds = [] (
      doc="""A list of id's to make invisible at the evaluation time.
      See \\ref UsdGeomPointInstancer_invisibleIds ."""
  )
}


class Camera "Camera" (
    doc = """Transformable camera.
    
    Describes optical properties of a camera via a common set of attributes
    that provide control over the camera's frustum as well as its depth of
    field. For stereo, the left and right camera are individual prims tagged
    through the \\ref UsdGeomCamera::GetStereoRoleAttr() "stereoRole attribute".
    
    There is a corresponding class GfCamera, which can hold the state of a
    camera (at a particular time). \\ref UsdGeomCamera::GetCamera() and
    \\ref UsdGeomCamera::SetFromCamera() convert between a USD camera prim and
    a GfCamera.

    To obtain the camera's location in world space, call the following on a
    UsdGeomCamera 'camera':
    \\code
    GfMatrix4d camXform = camera.ComputeLocalToWorldTransform(time);
    \\endcode
    \\note
    <b>Cameras in USD are always "Y up", regardless of the stage's orientation
    (i.e. UsdGeomGetStageUpAxis()).</b>  This means that the inverse of 
    'camXform' (the VIEW half of the <A HREF="http://www.glprogramming.com/red/chapter03.html#name2">MODELVIEW transform in OpenGL parlance</A>) 
    will transform the world such that the camera is at the origin, looking 
    down the -Z axis, with Y as the up axis.
    
    \\sa \\ref UsdGeom_LinAlgBasics
     """
    inherits = </Xformable>
    customData = {
        string extraIncludes = """
#include "pxr/base/gf/camera.h" """
    }
) {
    # viewing frustum
    token projection = "perspective" (
        allowedTokens = ["perspective", "orthographic"])
    float horizontalAperture  = 20.9550 (
        doc = """Horizontal aperture in millimeters (or, more general, tenths
                 of a world unit).
                 Defaults to the standard 35mm spherical projector aperture.""")
    float verticalAperture  = 15.2908 (
        doc = """Vertical aperture in millimeters (or, more general, tenths of
                 a world unit).
                 Defaults to the standard 35mm spherical projector aperture.""")
    float horizontalApertureOffset = 0.0 (
        doc = """Horizontal aperture offset in the same units as
                 horizontalAperture. Defaults to 0.""")
    float verticalApertureOffset = 0.0 (
        doc = """Vertical aperture offset in the same units as
                 verticalAperture. Defaults to 0.""")
    float focalLength = 50.0 (
        doc = """Perspective focal length in millimeters (or, more general,
                 tenths of a world unit).""")
    float2 clippingRange = (1, 1000000) (
        doc = """Near and far clipping distances in centimeters (or, more
                 general, world units).""")
    float4[] clippingPlanes = [] (
        doc = """Additional, arbitrarily oriented clipping planes.
                 A vector (a,b,c,d) encodes a clipping plane that cuts off
                 (x,y,z) with a * x + b * y + c * z + d * 1 < 0 where (x,y,z)
                 are the coordinates in the camera's space.""")

    # depth of field
    float fStop = 0.0 (
        doc = """Lens aperture. Defaults to 0.0, which turns off focusing.""")
    float focusDistance = 0.0 (
        doc = """Distance from the camera to the focus plane in centimeters (or
                 more general, world units).""")

    # stereoscopic 3D
    uniform token stereoRole = "mono" (
        allowedTokens = ["mono", "left", "right"]
        doc = """If different from mono, the camera is intended to be the left
                 or right camera of a stereo setup.""")

    # Parameters for motion blur
    double shutter:open = 0.0 (
        doc = """Frame relative shutter open time in UsdTimeCode units (negative
                 value indicates that the shutter opens before the current
                 frame time). Used for motion blur."""
    )
    double shutter:close = 0.0 (
        doc = """Frame relative shutter close time, analogous comments from
                 shutter:open apply. A value greater or equal to shutter:open
                 should be authored, otherwise there is no exposure and a
                 renderer should produce a black image."""
    )
}

class "GeomModelAPI"
(
    inherits = </APISchemaBase>
    customData = {
         string className = "ModelAPI"
         string extraIncludes = """
#include "pxr/usd/usdGeom/bboxCache.h"
#include "pxr/usd/usdGeom/constraintTarget.h"
#include "pxr/usd/usdGeom/imageable.h" """
    }
    doc = """UsdGeomModelAPI extends the generic UsdModelAPI schema with
    geometry specific concepts such as cached extents for the entire model,
    constraint targets, and geometry-inspired extensions to the payload
    lofting process.
   
    As described in GetExtentsHint() below, it is useful to cache extents
    at the model level.  UsdGeomModelAPI provides schema for computing and
    storing these cached extents, which can be consumed by UsdGeomBBoxCache to
    provide fast access to precomputed extents that will be used as the model's
    bounds ( see UsdGeomBBoxCache::UsdGeomBBoxCache() ).

    \\section UsdGeomModelAPI_drawMode Draw Modes

    Draw modes provide optional alternate imaging behavior for USD subtrees with
    kind model. \\em model:drawMode (which is inheritable) and
    \\em model:applyDrawMode (which is not) are resolved into a decision to stop
    traversing the scene graph at a certain point, and replace a USD subtree
    with proxy geometry.

    The value of \\em model:drawMode determines the type of proxy geometry:
    - \\em origin - Draw the model-space basis vectors of the replaced prim.
    - \\em bounds - Draw the model-space bounding box of the replaced prim.
    - \\em cards - Draw textured quads as a placeholder for the replaced prim.
    - \\em default - An explicit opinion to draw the USD subtree as normal.

    \\em model:drawMode is inheritable so that a whole scene, a large group, or
    all prototypes of a model hierarchy PointInstancer can be assigned a draw
    mode with a single attribute edit.  \\em model:applyDrawMode is meant to be
    written when an asset is authored, and provides flexibility for different
    asset types. For example, a character assembly (composed of character,
    clothes, etc) might have \\em model:applyDrawMode set at the top of the
    subtree so the whole group can be drawn as a single card object. An effects
    subtree might have \\em model:applyDrawMode set at a lower level so each
    particle group draws individually.

    Models of kind component are treated as if \\em model:applyDrawMode
    were true.  This means a prim is drawn with proxy geometry when: the
    prim has kind component, and/or \\em model:applyDrawMode is set; and
    the prim or an ancestor has a non-default value for \\em model:drawMode.
    A value for \\em model:drawMode on a child prim takes precedence over a
    value on a parent prim.

    \\section UsdGeomModelAPI_cardGeometry Cards Geometry

    The specific geometry used in cards mode is controlled by the
    \\em model:cardGeometry attribute:
    - \\em cross - Generate a quad normal to each basis direction and negative.
                   Locate each quad so that it bisects the model extents.
    - \\em box   - Generate a quad normal to each basis direction and negative.
                   Locate each quad on a face of the model extents, facing out.
    - \\em fromTexture - Generate a quad for each supplied texture from
                         attributes stored in that texture's metadata.

    For \\em cross and \\em box mode, the extents are calculated for purposes
    \\em default, \\em proxy, and \\em render, at their earliest authored time.
    If the model has no textures, all six card faces are rendered using
    \\em model:drawModeColor. If one or more textures are present, only axes
    with one or more textures assigned are drawn.  For each axis, if both
    textures (positive and negative) are specified, they'll be used on the
    corresponding card faces; if only one texture is specified, it will be
    mapped to the opposite card face after being flipped on the texture's
    s-axis. Any card faces with invalid asset paths will be drawn with
    \\em model:drawModeColor.

    Both \\em model:cardGeometry and \\em model:drawModeColor should be
    authored on the prim where the draw mode takes effect, since these
    attributes are not inherited.

    For \\em fromTexture mode, only card faces with valid textures assigned
    are drawn. The geometry is generated by pulling the \\em worldtoscreen
    attribute out of texture metadata.  This is expected to be a 4x4 matrix
    mapping the model-space position of the card quad to the clip-space quad
    with corners (-1,-1,0) and (1,1,0).  The card vertices are generated by
    transforming the clip-space corners by the inverse of \\em worldtoscreen.
    Textures are mapped so that (s) and (t) map to (+x) and (+y) in clip space.
    If the metadata cannot be read in the right format, or the matrix can't
    be inverted, the card face is not drawn.

    All card faces are drawn and textured as single-sided.

    \\todo CreatePayload() """
)
{
    uniform token model:drawMode (
        allowedTokens = ["origin", "bounds", "cards", "default"]
        doc = """Alternate imaging mode; applied to this prim or child prims
                 where \\em model:applyDrawMode is true, or where the prim
                 has kind \\em component. See \\ref UsdGeomModelAPI_drawMode
                 for mode descriptions."""
    )
    uniform bool model:applyDrawMode (
        doc = """If true, and this prim or parent prims have \\em model:drawMode
                 set, apply an alternate imaging mode to this prim. See
                 \\ref UsdGeomModelAPI_drawMode."""
    )
    uniform float3 model:drawModeColor (
        doc = """The base color of imaging prims inserted for alternate
                 imaging modes. For \\em origin and \\em bounds modes, this
                 controls line color; for \\em cards mode, this controls the
                 fallback quad color. If unspecified, it should be interpreted
                 as (0.18, 0.18, 0.18)."""
    )
    uniform token model:cardGeometry (
        allowedTokens = ["cross", "box", "fromTexture"]
        doc = """The geometry to generate for imaging prims inserted for \\em
                 cards imaging mode. See \\ref UsdGeomModelAPI_cardGeometry for
                 geometry descriptions. If unspecified, it should be interpreted
                 as \\em cross."""
    )
    asset model:cardTextureXPos (
        doc = """In \\em cards imaging mode, the texture applied to the X+ quad.
                 The texture axes (s,t) are mapped to model-space axes (-y, -z)."""
    )
    asset model:cardTextureYPos (
        doc = """In \\em cards imaging mode, the texture applied to the Y+ quad.
                 The texture axes (s,t) are mapped to model-space axes (x, -z)."""
    )
    asset model:cardTextureZPos (
        doc = """In \\em cards imaging mode, the texture applied to the Z+ quad.
                 The texture axes (s,t) are mapped to model-space axes (x, -y)."""
    )
    asset model:cardTextureXNeg (
        doc = """In \\em cards imaging mode, the texture applied to the X- quad.
                 The texture axes (s,t) are mapped to model-space axes (y, -z)."""
    )
    asset model:cardTextureYNeg (
        doc = """In \\em cards imaging mode, the texture applied to the Y- quad.
                 The texture axes (s,t) are mapped to model-space axes (-x, -z)."""
    )
    asset model:cardTextureZNeg (
        doc = """In \\em cards imaging mode, the texture applied to the Z- quad.
                 The texture axes (s,t) are mapped to model-space axes (-x, -y)."""
    )
}

class "MotionAPI"
(
    inherits = </APISchemaBase>
    doc = """UsdGeomMotionAPI encodes data that can live on any prim that
    may affect computations involving:
    - computed motion for motion blur
    - sampling for motion blur
    
    For example, UsdGeomMotionAPI provides *velocityScale* 
    (GetVelocityScaleAttr()) for controlling how motion-blur samples should
    be computed by velocity-consuming schemas."""
)
{
    float motion:velocityScale = 1.0 (
        customData = {
            string apiName = "velocityScale"
        }
        doc = """VelocityScale is an **inherited** float attribute that
        velocity-based schemas (e.g. PointBased, PointInstancer) can consume
        to compute interpolated positions and orientations by applying
        velocity and angularVelocity, which is required for interpolating 
        between samples when topology is varying over time.  Although these 
        quantities are generally physically computed by a simulator, sometimes 
        we require more or less motion-blur to achieve the desired look.  
        VelocityScale allows artists to dial-in, as a post-sim correction, 
        a scale factor to be applied to the velocity prior to computing 
        interpolated positions from it.
        
        See also ComputeVelocityScale()"""
    )
}