PartNet

PartNet Dataset Overview

Dataset Introduction

PartNet is a large‑scale 3D object understanding benchmark dataset comprising 26,671 3D models covering 24 object categories, with a total of 573,585 part instances. It supports tasks such as shape analysis, dynamic 3D scene modeling and simulation, functional analysis, and defines three benchmark tasks for evaluating 3D part recognition.

Dataset Structure

• stats/: Stores all valid PartNet annotation metadata, including annotation IDs, version IDs, categories, ShapeNet model IDs, annotator IDs, etc.
• scripts/: Contains scripts for data processing, such as JSON files for merging results and scripts for generating H5 instance‑segmentation files.
• data/: Holds downloaded PartNet data, including raw annotations, merged annotations, related metadata, and visualization files.

Data Visualization

PartNet provides two visualization pages:

• Original annotation visualization: http://download.cs.stanford.edu/orion/partnet_dataset/data_v0/42/tree_hier.html
• Merged annotation visualization: http://download.cs.stanford.edu/orion/partnet_dataset/data_v0/42/tree_hier_after_merging.html

Dataset Download

PartNet can be downloaded from the official ShapeNet website after registering as a user. Download link: https://www.shapenet.org/download/parts

Error Feedback

Users who discover annotation errors can report them through this form for correction in the next PartNet release.

Citation Information

@InProceedings{Mo_2019_CVPR, author = {Mo, Kaichun and Zhu, Shilin and Chang, Angel X. and Yi, Li and Tripathi, Subarna and Guibas, Leonidas J. and Su, Hao}, title = {{PartNet}: A Large‑Scale Benchmark for Fine‑Grained and Hierarchical Part‑Level {3D} Object Understanding}, booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)}, month = {June}, year = {2019} }

When using ShapeNet models, please also cite ShapeNet.