USD IO Customization Proposal #101283

New Issue

Michael Kowalski · 2022-09-23T01:38:15+02:00

Michael Kowalski commented

2022-09-23 01:38:15 +02:00

The following description was provided by @charlesfleche on the pipeline-assets-io-module Blender chat channel on 22 Sept 2022.

We want to support our custom metadata and attributes for a given prim, import and export. A few use cases:

ID to bind a Blender object to our game engine, so we can import in blender / edit / export to our database.
In games, skeletons are usually made of several smaller skeletons: one for the main body, but others for the weapon, clothes, accessories, etc... Those are stored as different objects in our game database, so you mix and match (think about those character creators where you choose the hair, the body, the eyes, the hat of your character...). We need to tag in USD the owner of each bone (with our custom schemas), store this info in Blender, and reserialize it to USD on the way back to the engine.
USD doesn't have a definite schema to define an object transform and its offset transform. We express those as two transforms ops in USD, but Blender can't make sense of those at the moment, so we should be able to post process an object to handle our custom schemas and apply those transforms.
Imagine a new object is created in Blender, then published to the game engine (we have a network API for that purpose). The response we get contains the new IDs of the objects in the Engine, and we need to assign those IDs to existing blender objects in the Blender scene.

So far we've seen two approaches to customize USD I/O in DCCs:

The Maya one, based on Schema translators. Here you register a schema and call some code when it is found while walking the Prims at import time.
Our own, based on hooks in the import / export loops:

def import_to_dcc(path):

    def recursively_create_node(prim, parent_dcc_node=None):
        if not prim.IsPseudoRoot():
            node = hook.create_node(prim=prim, parent_dcc_node=parent_dcc_node)            
            if node:
                hook.post_create_node(prim=prim, node=node)
                parent_dcc_node=node

        for child_prim in prim.GetChildren():
            recursively_create_nodes(child_prim, parent_dcc_node)

    # Import loop
    stage = hook.open_import_stage(path=path)
    hook.pre_stage_import(stage=stage)
    recursively_create_nodes(stage.GetPseudoRoot())
    hook.post_stage_import(stage=stage)

This loop does not do anything but to define hooks to be called at certain time during the lifecycle of an import:

open a stage
allow importers to initialize themselves after a stage is imported
recursively walk the USD tree to create prim
post process a prim (handle custom metadata, apply materials)
cleanup after the stage is imported

An implementation of a hook could look like this, for camera for example:

@hookimpl
def create_node(prim, parent_dcc_node):
"""Camera USD -> DCC importer hook implementation"""
    usd_camera = UsdGeom.Camera(prim)

    - Leave if the prim cannot be handled by this importer
    - Another importer might handle this prim, or it won't
    # be imported at all

    if not usd_camera:
        return None

    dcc_camera = build_dcc_camera()
    set_dcc_camera_from_usd_camera(usd_camera, dcc_camera)

Is the current prim to be imported a camera?
No ? I return None and let another importer deal with this prim.
Yes ? I create a new camera in my DCC and transfer the USD attributes to it. Other importers won't be called.

By the way, this stuff is based on pluggy .

I personally don't have a preference for the API style (hook based vs schema based). I've the feeling that being hook based allows for more flexibility, while Schema based could be more performant as it'd be faster to dispatch import code based on the presence of an Api (rather than the "is my UsdGeomCamera valid ?" of my hooks).

I've also seen in other DCCs that the creation of the prims happens at the Sdf level, not the Usd level, for performance reasons:

a first hook receives a layer, not a stage, to create a prim
another hook receives an UsdPrim so it can leverage the USD SchemaAPI system

That's quite neat, I think.

Another point to note is that for the DCCs I've seen (Maya and 3ds Max), those customizations are not DCC plugins, but USD plugins. The custom code is actually registered through the USD plugin system, not the DCC's.

The following description was provided by @charlesfleche on the `pipeline-assets-io-module` Blender chat channel on 22 Sept 2022. --- We want to support our custom metadata and attributes for a given prim, import and export. A few use cases: - ID to bind a Blender object to our game engine, so we can import in blender / edit / export to our database. - In games, skeletons are usually made of several smaller skeletons: one for the main body, but others for the weapon, clothes, accessories, etc... Those are stored as different objects in our game database, so you mix and match (think about those character creators where you choose the hair, the body, the eyes, the hat of your character...). We need to tag in USD the owner of each bone (with our custom schemas), store this info in Blender, and reserialize it to USD on the way back to the engine. - USD doesn't have a definite schema to define an object transform and its offset transform. We express those as two transforms ops in USD, but Blender can't make sense of those at the moment, so we should be able to post process an object to handle our custom schemas and apply those transforms. - Imagine a new object is created in Blender, then published to the game engine (we have a network API for that purpose). The response we get contains the new IDs of the objects in the Engine, and we need to assign those IDs to existing blender objects in the Blender scene. So far we've seen two approaches to customize USD I/O in DCCs: - The Maya one, based on [Schema translators](https://github.com/Autodesk/maya-usd/blob/dev/lib/mayaUsd/fileio/doc/SchemaAPI_Import_Export_in_Python.md). Here you register a schema and call some code when it is found while walking the Prims at import time. - Our own, based on hooks in the import / export loops: ``` def import_to_dcc(path): def recursively_create_node(prim, parent_dcc_node=None): if not prim.IsPseudoRoot(): node = hook.create_node(prim=prim, parent_dcc_node=parent_dcc_node) if node: hook.post_create_node(prim=prim, node=node) parent_dcc_node=node for child_prim in prim.GetChildren(): recursively_create_nodes(child_prim, parent_dcc_node) # Import loop stage = hook.open_import_stage(path=path) hook.pre_stage_import(stage=stage) recursively_create_nodes(stage.GetPseudoRoot()) hook.post_stage_import(stage=stage) ``` This loop does not do anything but to define hooks to be called at certain time during the lifecycle of an import: - open a stage - allow importers to initialize themselves after a stage is imported - recursively walk the USD tree to create prim - post process a prim (handle custom metadata, apply materials) - cleanup after the stage is imported An implementation of a hook could look like this, for camera for example: ``` @hookimpl def create_node(prim, parent_dcc_node): """Camera USD -> DCC importer hook implementation""" usd_camera = UsdGeom.Camera(prim) - Leave if the prim cannot be handled by this importer - Another importer might handle this prim, or it won't # be imported at all if not usd_camera: return None dcc_camera = build_dcc_camera() set_dcc_camera_from_usd_camera(usd_camera, dcc_camera) ``` - Is the current prim to be imported a camera? - No ? I return None and let another importer deal with this prim. - Yes ? I create a new camera in my DCC and transfer the USD attributes to it. Other importers won't be called. By the way, this stuff is based on [pluggy ](https://pluggy.readthedocs.io/en/stable/). I personally don't have a preference for the API style (hook based vs schema based). I've the feeling that being hook based allows for more flexibility, while Schema based could be more performant as it'd be faster to dispatch import code based on the presence of an Api (rather than the "is my `UsdGeomCamera` valid ?" of my hooks). I've also seen in other DCCs that the creation of the prims happens at the `Sdf` level, not the `Usd` level, for performance reasons: - a first hook receives a layer, not a stage, to create a prim - another hook receives an `UsdPrim` so it can leverage the `USD SchemaAPI` system That's quite neat, I think. Another point to note is that for the DCCs I've seen (Maya and 3ds Max), those customizations **are not DCC plugins**, but **USD plugins**. The custom code is actually registered through the USD plugin system, not the DCC's.

Michael Kowalski commented

2022-09-23 01:38:15 +02:00

Added subscribers: @charlesfleche, @makowalski