3D Dataloader: ZIP Loader#

We give here a few pointers on how to easily sample raw, stylized meshes and pointclouds from our zipped 3D models.

Note that the ZIP loader enables to sample up to a 1000 average compositions (styles) per model, by texturing meshes β€œon-the-fly” following our generated style metadata.

from demo_utils_3D import *
import utils3D.plot as plt_utils

METADATA_DIR = "./metadata/"
ZIP_PATH = "./3D_data/3DCoMPaT_ZIP.zip"

Working with raw meshes#

To load unstylized 3D shapes, we provide the ShapeLoader class. The n_points parameter enables sampling of pointclouds when loading the shapes.

It can be instantiated with the load_mesh parameter toggled to interactively visualize the shapes with a placeholder style.

Meshes fetched by the loader are trimesh.Trimesh instances that contain all relevant geometry information.

"""Unsylized 3D shape loader.

    zip_path:        3DCoMPaT models zip directory.
    meta_dir:        Metadata directory.
    split:           One of {train, valid}.
    semantic_level:  Semantic level to use for segmentations. One of {fine, medium, coarse}
    n_points:        Number of sampled points.
    load_mesh:       Only load meshes.
    shape_only:      Ignore part segments while sampling pointclouds.
    get_normals:     Also return normal vectors for each sampled point.
    shuffle:         Shuffle the dataset.
    seed:            Initial random seed.
from compat3D import ShapeLoader

train_dataset = ShapeLoader(zip_path=ZIP_PATH,

# Iterating over the loader
for shape_id, shape_label, mesh in train_dataset: