Repetitive Patterns on Textured 3D Surfaces

We present an annotated dataset of 82 different 3D models of painted ancient Perivial vessels, exhibiting different levels of repetitiveness in their surface patterns. Alongside the data, we provide the tools used to obtain the annotations as well as a benchmark that can rank different recognition techniques regarding their applicability for this dataset. More details are given in the corresponding publication A Benchmark Dataset for Repetitive Pattern Recognition on Textured 3D Surfaces submitted to the Symposium on Geometry Processing 2021.


The basis of the dataset is made up of 82 textured 3D meshes, captured with a structured-light scanner by the Josefina Ramos de Cox museum in Lima, Perú. The collection comprises several shape classes, such as jars, pitchers, bowls,figurines, basins, pots, plates, and vases. Moreover, the objects are attributed to several pre-Columbian cultures, each with their own characteristic artistic styles in shapes and paintings. The models have been manually post-processed and contain about 130K triangles per model.

From the patterns exhibited by the object surfaces characteristic pattern archetypes have been identified by archaeologists before each of their occurrces was annotated on a per-face basis with a specialized tool.


We provide the annotated dataset, published under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International licence, together with the source code for the tools used to obtained the annotations as well as for conducting the evaluation.


The whole dataset is available as a zip file or gzipped tar file. Each contain one subfolder per model including
  • The original model after preprocessing in PLY format (<model_id>.ply) with its texture map (<model_id>.jpg),
  • One or more of the pattern archetypes (<model_id>.pat<pattern_id>.png),
  • A high-resolution RGBA image of the flattened surface (<model_id>_flat.png),
  • The face-based annotation in JSON format (<model_id>.json) and
  • A thumbnail of the flattened annotated surface (<model_id>_thumb.jpg).
Release Date Version SHA256 Checksum Download
2021-06-15 0.1 04c3905cae052d21a5042774d895f185551e6ecf44a687d56c6bd8058ec636fb
2021-06-15 0.1 01dfd2941bccf1ded7803854f4c70edeb7935908a1db860ae6213bc01e21b008 pattern-benchmark-v0.1.tar.gz

Annotation Tools

The annotations have been obtained with two interoperating tools. The first tool Sample Pattern Selection is responsible for determining distinct similarity classes while the second tool Pattern Entity Selection was empolyed to annotated individual occurences on the object surface. The sources for both tools are available at and respectively.

Evaluation Tools



For citing the benchmark please refer to [Lengauer et al. 2021]:
@article {10.1111:cgf.14352,
    journal = {Computer Graphics Forum},
    title = {{A Benchmark Dataset for Repetitive Pattern Recognition on Textured 3D Surfaces}},
    author = {Lengauer, Stefan and Sipiran, Ivan and Preiner, Reinhold and Schreck, Tobias and Bustos, Benjamin},
    year = {2021},
    publisher = {The Eurographics Association and John Wiley & Sons Ltd.},
    ISSN = {1467-8659},
    DOI = {10.1111/cgf.14352}


In case you detect issues with the data or source code, want to provide feedback on the benchmark or have data yourself which you want to contribute to the benchmark please let us know via .

This work was co-funded by the Austrian Science Fund FWF and the State of Styria, Austria within the project Crossmodal Search and Visual Exploration of 3D Cultural Heritage Objects (P31317-NBL). This work has been also partially supported by Proyecto de Mejoramiento y Ampliación de los Servicios del Sistema Nacional de Ciencia, Tecnología e Innovación Tecnológica (Banco Mundial, Concytec), Nr. Grant 062-2018-FONDECYT-BM-IADT-AV and also by the ANID - Millennium Science Initiative Program - Code ICN17_002.
Last updated: July 20, 2021