A Large-Scale Shape Benchmark for 3D Object Retrieval

Toyohashi Shape Benchmark


The Toyohashi Shape Benchmark (TSB) are labeled subsets of the NTU 3D Model Database and the SHREC 2010 Generic Dataset. The TSB consists of 10,000 3D shapes with 352 categories.

For the collecting 3D models, we employ publicly available data, which have not yet been labeled. Specifically, a part of our 3D shape models come from the NTU 3D Model Database (NTU) [1], which consists of 10,910 unclassified and unlabeled models. The problem is that most of them are extremely difficult for humans to classify and label. Another part of our models come from 3,168 models of SHREC’10 Track: Generic 3D Warehouse [2]. These two parts total 14,078 unclassified and unlabeled models.

Our laboratory members joined to work for manually classifying the 3D shape data collected. We repeated the process of the category construction by carefully examining its validity, and ended up with 352 categories for 10,000 3D shape models.



The TSB can only be used for academic purposes and cannot be used for commercial products.



In citing the Toyohashi Shape Benchmark, please refer to:

Atsushi Tatsuma, Hitoshi Koyanagi, and Masaki Aono, "A Large-Scale Shape Benchmark for 3D Object Retrieval: Toyohashi Shape Benchmark," In Proc. of APSIPA2012 (Asia Pacific Signal and Information Processing Association), Hollywood Califolnia, USA, 2012.


[1] NTU 3D Model Database ver.1

[2] SHREC 2010 - Shape Retrieval Contest based on Generic 3D Warehouse

[3] Knowledge Data Engineering and Information Retrieval Laboratory