05 — COMPUTATIONAL HERITAGE RESEARCH

05 — COMPUTATIONAL HERITAGE RESEARCH

Heritage Point-Cloud Segmentation

Heritage Point-Cloud Segmentation

Heritage Point-Cloud Segmentation

A computational heritage research project investigating how photogrammetric data can be enhanced and structured for machine-learning-based segmentation of architectural surfaces. The workflow combines UV mapping, representative geometry, and data preprocessing to support unit-level analysis of historic masonry.

A computational heritage research project investigating how photogrammetric data can be enhanced and structured for machine-learning-based segmentation of architectural surfaces. The workflow combines UV mapping, representative geometry, and data preprocessing to support unit-level analysis of historic masonry.

A computational heritage research project investigating how photogrammetric data can be enhanced and structured for machine-learning-based segmentation of architectural surfaces. The workflow combines UV mapping, representative geometry, and data preprocessing to support unit-level analysis of historic masonry.

RESEARCH TEAM

RESEARCH TEAM

Fatih Uzun · Sevgi Altun · Sena Kayasü · Berkay Öztürk · Yusuf Şahin · Gözde Ünal · Mine Özkar

Fatih Uzun · Sevgi Altun · Sena Kayasü · Berkay Öztürk · Yusuf Şahin · Gözde Ünal · Mine Özkar

Role
Researcher · Computational Workflow Development

Role
Researcher · Computational Workflow Development

Role
Researcher · Computational Workflow Development

Research Methods
Photogrammetry · Point-Cloud Processing · Machine-Learning Evaluation

Research Methods
Photogrammetry · Point-Cloud Processing · Machine-Learning Evaluation

Research Methods
Photogrammetry · Point-Cloud Processing · Machine-Learning Evaluation

Tools
Blender · UV Mapping · PyTorch · DGCNN · CurveNet

Tools
Blender · UV Mapping · PyTorch · DGCNN · CurveNet

Tools
Blender · UV Mapping · PyTorch · DGCNN · CurveNet

FUNDED RESEARCH

TÜBİTAK 1001 · 2023

FUNDED RESEARCH

TÜBİTAK 1001 · 2023

FUNDED RESEARCH

TÜBİTAK 1001 · 2023

01 — FROM PHOTOGRAMMETRY TO STRUCTURED DATA

01 — FROM PHOTOGRAMMETRY TO STRUCTURED DATA

01 — FROM PHOTOGRAMMETRY TO STRUCTURED DATA

Photogrammetric models provide detailed geometric and visual information, but their raw point-cloud data does not directly express the semantic structure of architectural components. This study investigates how that data can be reorganised and enhanced before machine-learning-based segmentation.


The workflow focuses on the glazed-brick dome of the 13th-century Sahip Ata Hanikah in Konya, using a photogrammetric high-poly model as the starting point for unit-level analysis.

02 — UV-BASED PREPROCESSING

02 — UV-BASED PREPROCESSING

02 — UV-BASED PREPROCESSING

A representative low-poly model was aligned with the photogrammetric source model and given a regular UV layout. Colour and surface information were then transferred through baking, masking, and displacement-map generation to create structured data that could be used in subsequent segmentation tasks.


This preprocessing step made the surface geometry and brick patterns more operable while reducing the computational complexity of the original model.

Figure — High-poly and representative low-poly UV mappings used in the preprocessing workflow.

Figure — High-poly and representative low-poly UV mappings used in the preprocessing workflow.

Figure — High-poly and representative low-poly UV mappings used in the preprocessing workflow.

03 — MACHINE-LEARNING SEGMENTATION

03 — MACHINE-LEARNING SEGMENTATION

03 — MACHINE-LEARNING SEGMENTATION

The prepared data was evaluated using DGCNN and CurveNet point-cloud segmentation models. The dome was classified into four categories: mortar, unglazed brick, turquoise-glazed brick, and black-glazed brick.


Comparisons between datasets with and without UV-based preprocessing showed small but measurable improvements in segmentation accuracy, indicating that the proposed workflow can support the preparation of architectural heritage data for machine-learning applications.

With UV

With UV

With UV

Without UV

Without UV

Without UV

Change

Change

Change

DGCNN · mIoU

DGCNN · mIoU

DGCNN ·

mIoU

72.49

72.49

72.49

71.22

71.22

71.22

+1.27

+1.27

+1.27

DGCNN · Point Accuracy

DGCNN ·

Point Accuracy

DGCNN ·

Point

Accuracy

91.62

91.62

91.62

90.56

90.56

90.56

+1.06

+1.06

+1.06

CurveNet · mIoU

CurveNet · mIoU

CurveNet ·

mIoU

68.21

68.21

68.21

67.64

67.64

67.64

+0.57

+0.57

+0.57

CurveNet · Point Accuracy

CurveNet ·

Point Accuracy

CurveNet ·

Point

Accuracy

77.46

77.46

77.46

75.65

75.65

75.65

+1.81

+1.81

+1.81

04 — Publications & Research Outputs

04 — Publications & Research Outputs

04 — Publications & Research Outputs

This study was published as part of the CAADRIA 2024 proceedings, presenting the proposed UV-based preprocessing workflow and its evaluation through machine-learning-based point-cloud segmentation.

This study was published as part of the CAADRIA 2024 proceedings, presenting the proposed UV-based preprocessing workflow and its evaluation through machine-learning-based point-cloud segmentation.

This study was published as part of the CAADRIA 2024 proceedings, presenting the proposed UV-based preprocessing workflow and its evaluation through machine-learning-based point-cloud segmentation.

PUBLISHED · 2024

PUBLISHED · 2024

Utilising UV-Mapping for the 3D-Point Cloud Segmentation of Architectural Heritage

Fatih Uzun · Sevgi Altun · Sena Kayasü · Berkay Öztürk · Yusuf Şahin · Gözde Ünal · Mine Özkar

CAADRIA 2024 · Accelerated Design

Read Paper ↗

Immersive Cultural Heritage Conservation Education: A VR-Based Serious Game on Stone Surface Deterioration


Fatih Uzun · Mine Özkar


eCAADe 2026 · Digital Heritage & Preservation

Design, technical implementation, and preliminary expert-centered evaluation of the immersive VR version of SurfaceQuest.

Read Paper ↗

FUNDING & RESEARCH CONTEXT

FUNDING & RESEARCH CONTEXT

FUNDING & RESEARCH CONTEXT

FUNDED RESEARCH · 2023

FUNDED RESEARCH · 2023

TÜBİTAK 1001 Research Project

Developing Computational Methods for Modeling Building Components Using Photogrammetry Data in the Documentation of Historic Buildings

PI: Mine Özkar

Project Scholar: Fatih Uzun

Project No. 119K896

TÜBİTAK 1001 Research Project

Developing Computational Methods for Modeling Building Components Using Photogrammetry Data in the Documentation of Historic Buildings

PI: Mine Özkar

Project Scholar: Fatih Uzun

Project No. 119K896

Publication details →