Cluster3D: A Dataset and Benchmark for Clustering Non-Categorical 3D CAD Models

Siyuan Xiang, Ching Tseng, Congcong Wen, Deshana Desai, Yifeng Kou, Binil Starly, Daniele Panozzo, Chen Feng

Keywords: 3D CAD clustering, deep clustering, non-categorical, CAD models, ABC

TL;DR: Cluster3D fills the gap in 3D shape clustering of non-categorical objects by annotating hundreds of thousands of pairwise similarities

Abstract: We introduce the first large-scale dataset and benchmark for non-categorical annotation and clustering of 3D CAD models. We use the geometric data of the ABC dataset, and we develop an interface to allow expert mechanical engineers to efficiently annotate pairwise CAD model similarities, which we use to evaluate the performance of seven baseline deep clustering methods. Our dataset contains a manually annotated subset of 22,968 shapes, and 252,648 annotations. Our dataset is the first to directly target deep clustering algorithms for geometric shapes, and we believe it will be an important building block to analyze and utilize the massive 3D shape collections that are starting to appear in deep geometric computing. Our results suggest that, differently from the already mature shape classification algorithms, deep clustering algorithms for 3D CAD models are in their infancy and there is much room for improving their performance.

URL: https://cluster3d.github.io/