Skip to content

Latest commit

 

History

History
109 lines (72 loc) · 4.5 KB

shapenet.md

File metadata and controls

109 lines (72 loc) · 4.5 KB
Raw
3D Model Libraries

Add ShapeNet models to TDW

You can add ShapeNet models to TDW by converting them to asset bundles.

ShapeNet is a huge repository of semantically tagged obj files available to users that has several advantages and disadvantages compared to TDW's models.

Advantages:

  • There are far more ShapeNet models and semantic categories than TDW's model libraries.
  • ShapeNet is free (as opposed to TDW's largest model library, which is non-free).

Disadvantages:

  • ShapeNet models are inferior to TDW's models.
    • Most of them are not photorealistic.
    • Some of them are corrupted or have features that make them unsuitable for TDW such as being a 2D quad.
    • Many of them have different "up" directions.
    • ShapeNet models were often created at different canonical scales; for example, two ShapeNet that are ostensibly 1 cubic meter may be totally different sizes when added to TDW.
  • ShapeNet models are not hosted by default as TDW asset bundles. This is because of the reasons listed above (it would be extremely time-consuming for us to vet all of ShapeNet) and for potential licensing reasons.

That said, you can generate your own ShapeNet asset bundles and host them locally.

Setup

ShapeNetCore

  1. Clone the tdw repo in order to use shapenet.py.
  2. See the requirements for using the ModelCreator.
  3. Download and extract ShapeNetCore.v2.zip.

ShapeNet SEM

  1. Clone the tdw repo in order to use shapenet.py.
  2. See the requirements for using the ModelCreator.
  3. Create a root directory, for example D:/shapenet_sem/
  4. Download and extract models-OBJ.zip into the root directory.
  5. Download and extract models-textures.zip into the root directory.
  6. Move every file in textures/ into models/.
  7. Download metadata.csv to the root directory.

Result:

D:/shapenet_sem/
....models/
....metadata.csv

Usge

  1. cd tdw/Python/shapenet
  2. python3 shapenet.py [ARGUMENTS]
Argument Type Description Default
--src str The absolute path to the root source directory. "D:/ShapeNetCore.v2"
--dest str The absolute path to the root destination directory. "D:/tdw_shapenet_core"
--set str Which ShapeNet set this is.
Choices: "core" or "sem"		 "core"
--vhacd_resolution int Higher value=better-fitting colliders and slower build process. 8000000

What the script will do

  1. Convert ShapeNet metadata into a new metadata.csv file: output_directory/metadata.csv.
  2. Run AssetBundleCreator.metadata_file_to_asset_bundles(). This will generate asset bundles and metadata records.

Time required

This is a very slow process. It can take multiple weeks to process all of the models.

To speed up the process, set --vhacd_resolution to a low value (e.g. 1000). This will make the physics colliders fit the model poorly, but will vastly speed up the overall asset bundle creation process. Even then, expect the whole process to require several days.

The process can be paused, restarted, etc. Once you've generated the initial library file, the script will scan for existing asset bundles and won't try to re-create them

How to use the models in TDW

Once you've created the asset bundles, you can add them to TDW like any other asset bundles:

from tdw.controller import Controller
from tdw.tdw_utils import TDWUtils

shapenet_library_path = "path/to/shapenet/records.json"

c = Controller()
model_librarian = ModelLibrarian(library=shapenet_library_path)
c.communicate([TDWUtils.create_empty_room(12, 12),
               c.get_add_object(model_name=model_librarian.records[0].name,
                                object_id=c.get_unique_id(),
                                library=shapenet_library_path)])

This is the last document in the "Non-Default 3D models" tutorial.

Return to the README

ShapeNet convertor script:

Python API: