PartNet Shape Archive with Part Annotations

PartNet Archive contains 3D object data with part-level annotations, derived from the ShapeNet repository. The prerelease v0 from March 2019 includes meshes, point clouds, and HDF5 files for semantic and instance segmentation tasks. ShapeNet assembled this collection for research in fine-grained 3D shape understanding.

Use Cases

• Train semantic segmentation models using labeled point clouds from the 'sem_seg_h5.zip' files.
• Develop instance segmentation algorithms on 3D shapes with part annotations referenced in the 'ins_seg_h5.zip' archive.
• Benchmark geometric deep learning methods on the provided meshes and point cloud chunks ('data_v0_chunk.zip').
• Analyze hierarchical part structures for object categories using the annotated PartNet v0 data.

Strengths

• Includes multiple data representations: annotated meshes, point clouds, and preprocessed HDF5 files.
• Dataset is derived from the established ShapeNet repository, a standard in 3D object research.
• Provides separate archives for semantic segmentation (8GB) and instance segmentation tasks.

Limitations

• Specific row counts, total object instances, and detailed class distributions are not provided in the description.
• Data is from a 2019 prerelease (v0), which may not reflect the most current annotations or object categories.
• The archive structure involves multiple large chunk files (10GB each), complicating download and extraction.

Provenance

Source

ShapeNet

Collection Method

Derived and annotated from the ShapeNet subset; specific annotation methodology is detailed in the PartNet paper.

Freshness

Last updated on the platform was 2024-03-22, but the core data is from the 2019 prerelease.