3DITA: 3D Indian Temple Architecture Dataset

Dataset Description

3DITA is India's first large-scale, benchmark point cloud dataset designed specifically for semantic segmentation of Nagara-style Indian temple architecture. This dataset comprises over 325 million manually annotated 3D points collected from 47 temple structures across 13 sites in Uttarakhand, India. While Indian temple architecture encompasses a vast array of styles, this dataset focuses on Nagara temples for the purpose of simplifying the workflow and showcasing the potential of advanced technologies in heritage conservation. The dataset captures key architectural components of Nagara-style temples, including walls, mandapa, shikhara, amalaka, and garbhagriha, offering rich geometric and semantic details essential for heritage documentation, reconstruction, and AI-driven analysis.

Acquisition Methods

The dataset was created using proximal remote sensing techniques:

• Close-Range Photogrammetry (CRP): 22,370 overlapping photographs collected from 46 temples were processed using Structure-from-Motion techniques to generate dense 3D point clouds.

• Terrestrial Laser Scanning (TLS): One temple site was scanned using a FARO Focus 350 scanner, capturing approximately 13 million points through 15 high-precision scans.
  The collected data was manually screened, aligned, and co-registered, resulting in high-resolution, geometrically accurate point clouds.

Annotation Format and Legend

• Point clouds are annotated manually with scalar labels for semantic classes, stored in LAS/ASCII formats then converted into '.txt' format for applying Deep Learning semantic segmentation algorithm and for data dissemination.

• Each 3D point is described by ten attributes, including its spatial coordinates (x, y, z), color values (red, green, blue), surface normals (Nx, Ny, Nz), and a class label that was assigned during preprocessing based on the filename.

• Indian temple architecture encompasses a vast array of styles and classifications, each with its unique features. The authors of this dataset acknowledge the diversity of Indian architecture and respect the wide range of architectural forms present. However, for the purpose of simplification and to efficiently frame a workflow for identifying the potential of advanced technologies in the conservation and protection of Indian architecture, this work focuses on the basic classification of Nagara-style temples. The following semantic classes have been defined for this purpose, and it is important to note that this classification is not intended to diminish the richness of other architectural traditions nor to hurt the sentiments of others.

  Semantic Classes:

  • Mandapa: Pillared hall, with the roof considered as part of the mandapa.

  • Walls: Structural boundaries constructed from stone.

  • Shikhara: Curvilinear tower above the inner sanctum.

  • Garbhagriha: The innermost sanctum; main deity chamber.

  • Amalaka: Circular stone disc crowning the temple apex.

Citation Information

If you use this dataset in your research, please cite the following article:

Sridhar, M., Paygude, A., Pande, H., & Tiwari, P. S. (2025). 3DITA: A 3D Benchmark Dataset for Nagara-Style Indian Temple Architecture. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XLVIII-M-9, 1435-1442. https://doi.org/10.5194/isprs-archives-XLVIII-M-9-2025-1435-2025

Licensing and Usage Notes

• The dataset is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

• It is made available for academic and non-commercial research use.

• Redistribution and derivative works are allowed provided proper citation.

• Commercial use requires prior permission from the data owners.

• Users are encouraged to acknowledge the Archaeological Survey of India (ASI) and Indian Institute of Remote Sensing (IIRS), ISRO for their support in data collection and research facilitation.