The Shape COSEG Dataset



December 8, 2012: we fixed several problems in the ground-truth of our modeled irons, tele-aliens, and chairs.


The goal of this work is to provide data for quantitative analysis of how people consistently segment a set of shapes and for evaluation of our active co-analysis algorithm. To build the dataset, we have collected 11 sets of shapes which possess a consistent ground-truth segmentation and labeling. Among them, seven sets from the dataset of Sidi et al. [2011]. Since we consider the labeling of large sets as one of the main motivations of our work, we created three additional large sets: tele-alines, vases and chairs. These three sets consists of 200, 300, 400 shapes, respetively. We also created a small but challenging set of irons.


All data can be downloaded for free via the following links.

8 small sets

Candelabra 28 Shapes Ground-truth
Chairs 20 Shapes Ground-truth
Fourleg 20 Shapes Ground-truth
Goblets 12 Shapes Ground-truth
Guitars 44 Shapes Ground-truth
Lamps 20 Shapes Ground-truth
Vases 28 Shapes Ground-truth
Irons 18 Shapes Ground-truth

3 large sets

Tele-aliens 200 Shapes Ground-truth
Vases 300 Shapes Ground-truth
Chairs 400 Shapes Ground-truth


The related projects from which the dataset grew to its current form are:

  • Yunhai Wang, Shmulik Asafi, Oliver van Kaick, Hao Zhang, Daniel Cohen-Or, Baoquan Chen, Active Co-Analysis of a Set of Shapes, ACM Transactions on Graphics (Proc. SIGGRAPH Asia), vol. 31, n. 6, 2012.
  • Oana Sidi, Oliver van Kaick, Yanir Kleiman, Hao Zhang, Daniel Cohen-Or, Unsupervised Co-Segmentation of a Set of Shapes via Descriptor-Space Spectral Clustering, ACM Transactions on Graphics (Proc. SIGGRAPH Asia), vol. 30, n. 6, pp. 126-134, 2011
  • Oliver van Kaick, Andrea Tagliasacchi, Oana Sidi, Hao Zhang, Daniel Cohen-Or, Lior Wolf, Ghassan Hamarneh, Prior Knowledge for Part Correspondence, Computer Graphics Forum (Proc. Eurographics), vol. 30, n. 2, pp. 553-562, 2011.

  • Credits

    We thank Xiaobai Chen (Princeton mesh segmenation bechmark), Daniela Giorgi and AIM@SHAPE (SHREC 2007 Watertight Track), and Ran Gal for some of the models used in the benchmark. Andrea Tagliasacchi, Oliver van Kaick, Oana Sidi, Yanir Kleiman, Shmulik Asafi, Yunhai Wang and the artists in SIAT: Jiacheng Ren and Guangfan Pan, have contributed to creating the other shapes.


    Projective shape analysis (PSA) dataset


    Please send email to cloudseawang [at] if you have any questions.