[Paper Review] ShapeNet: An Information-Rich 3D Model Repository

Anne Lynn S. Chang, Thomas Funkhouser+11 more|arXiv (Cornell University)|Dec 9, 2015

Image Retrieval and Classification TechniquesComputer Science2,362 citations

TL;DR

ShapeNet is a large-scale, semantically annotated repository of 3D CAD models organized by WordNet synsets, with rich geometric, functional, and physical annotations plus a public search/download interface.

ABSTRACT

We present ShapeNet: a richly-annotated, large-scale repository of shapes represented by 3D CAD models of objects. ShapeNet contains 3D models from a multitude of semantic categories and organizes them under the WordNet taxonomy. It is a collection of datasets providing many semantic annotations for each 3D model such as consistent rigid alignments, parts and bilateral symmetry planes, physical sizes, keywords, as well as other planned annotations. Annotations are made available through a public web-based interface to enable data visualization of object attributes, promote data-driven geometric analysis, and provide a large-scale quantitative benchmark for research in computer graphics and vision. At the time of this technical report, ShapeNet has indexed more than 3,000,000 models, 220,000 models out of which are classified into 3,135 categories (WordNet synsets). In this report we describe the ShapeNet effort as a whole, provide details for all currently available datasets, and summarize future plans.

Motivation & Objective

• Provide a large-scale, semantically enriched collection of 3D models for research in graphics and vision.
• Annotate models with orientations, parts, symmetries, and physical attributes to enable data-driven analysis.
• Link 3D models to WordNet and other modalities (e.g., ImageNet) to enable cross-domain research.
• Offer a web-based interface and data access tools for searching, visualization, and benchmarking.
• Support ongoing expansion with additional annotation types and correspondences across data sources.

Proposed method

• Collect 3D models from public repositories such as Trimble 3D Warehouse and Yobi3D.
• Organize models under WordNet synsets to form a hierarchical taxonomy view.
• Annotate models with rigid alignments, parts and keypoints, symmetries, and object sizes.
• Estimate and annotate physical properties including surface materials and weights.
• Use a hybrid annotation approach that combines algorithmic predictions with crowd-sourcing and expert verification.
• Provide a Solr-based index and web API for searching, filtering, and batch downloading.

Experimental results

Research questions

• RQ1 How can a large-scale 3D model repository be organized semantically to support data-driven research in graphics and vision?
• RQ2 What annotations (geometric, functional, physical) are most valuable for downstream tasks such as segmentation, alignment, and recognition?
• RQ3 How can correspondences and links between 3D models and other modalities (e.g., images) be established and leveraged?
• RQ4 What is the scale and category distribution of a richly annotated ShapeNet dataset, and how does it compare to prior 3D model collections?

Key findings

• ShapeNet indexes roughly 3,000,000 models, with 220,000 models classified into 3,135 WordNet synsets.
• ShapeNetCore subset covers 55 common object categories and about 51,300 models.
• ShapeNetSem subset contains 12,000 models across 270 categories with additional real-world dimension and volume annotations.
• ShapeNet provides a dense network of annotations including upright/front orientation, parts, keypoints, symmetries, and hierarchical part decompositions.
• An initial data collection relies on public repositories (Trimble 3D Warehouse and Yobi3D) and uses a hybrid annotation strategy combining algorithmic predictions with human verification.
• The project uses a Solr-based index and offers a web interface and batched downloads for researchers.