Monday, June 10, 2013

A 3D Pottery Content Based Retrieval Benchmark Dataset

About the 3D Pottery Benchmark


The existence of benchmark data sets is essential in order to perform the evaluation and the comparison of content based retrieval methods. Although, no standard benchmark is available for 3D data shape matching there is currently a number of 3D data sets have been proposed by several research teams and are freely offered to be used.In order to evaluate our pottery specific 3D shape descriptors, we had to create a dataset of polygonal 3D vessels. The current pottery dataset is composed by a total number of 1012 digitised, manually modelled and semiautomated generated 3D models (3D vessel random generator). The content of the dataset is classified in several generic vessel shape categories such as Ancient Greek (Alabastron, Amphora, Hydria, Kantharos, Lekythos, Psykter, etc), Native American (Jar, Effigy, Bowl, Bottle, etc), modern pottery and others. The current classification has been performed with the help of the Department of Cultural Heritage of the ILSP/Athena Research Centre at an archaeologist-oriented semantic level. Please note that the current classification isn't the final one and we will be soon presenting a shape-oriented one. Furthermore, the 3D models are stored in the Wavefront OBJ file format and are currently followed by a single view thumbnail in JPEG file format and no texture information. A text file and an excel file is also provided where the filename of each model is assigned to a class. This information can be used for the computation of any performance metrics.

Read More:

No comments: