HOTOSM-3D-TT

This repository contains pipelines for processing large-scale 3D datasets.

The workflow is divided into three main components:

1. Instance Segmentation – 3D building instance segmentation using SoftGroup (separate individual buildings)
2. Semantic Segmentation & Clustering – Point cloud classification (ground, building, vegetation) and clustering method to filter building points and detect individual buildings
3. Building Height Estimation – Extract building heights from delineated point clouds

---

📂 Repository Structure

Note: This repository is currently organized as separate branches. Each branch contains one module.

instance_segmentation_softgroup/        # Branch for instance segmentation
├── SoftGroup/
│   ├── configs/softgroup/
│   │   └── softgroup_clientdata.yaml   # Configuration for client data
│   ├── dataset/clientdata/
│   │   └── prepare_data_clientdata.py  # Custom preprocessing
│   ├── docs/
│   │   ├── INSTALLATION_GUIDE.md       # Installation instructions
│   │   ├── RUNPOD_SETUP_GUIDE.md       # Cloud setup guide
│   │   └── ...
│   ├── softgroup/                      # Core model implementation
│   ├── tools/                          # Training & inference scripts
│   └── requirements.txt
└── README.md

semantic_segmentation_and_clustering/   # Branch for semantic segmentation & clustering
├── .gitignore
├── README.md
├── requirements.txt
├── Final models/                       # Trained model checkpoints
├── building_delineation/
│   ├── clustering/                     # DBSCAN, HDBSCAN clustering algorithms
│   ├── configs/                        # Clustering configuration
│   ├── evaluation/                     # Validation and visualization
│   ├── notebooks/                      # Demo notebooks
│   ├── README.md
│   └── requirements.txt
└── semantic_segmentation/
    ├── configs/                        # Segmentation configuration
    ├── data_preprocessing/             # Label remapping, dataset preparation
    ├── training/                       # Training on STPLS3D, Semantic3D
    ├── validation/                     # Evaluation scripts
    ├── notebooks/                      # Experiments and analysis
    ├── README.md
    └── requirements.txt

building_height/                        # Branch for height estimation
├── configs/
│   └── height_config.yaml              # Height calculation configuration
├── calculate_building_heights.py       # Main height calculation script
├── compare_gt_vs_inference.py          # Ground truth vs inference comparison
├── requirements.txt
└── README.md

---

🌿 Branch Access

Module	Branch Name	Documentation
Instance Segmentation	instance_segmentation_softgroup	See README in branch
Semantic Segmentation & Clustering	semantic_segmentation_and_clustering	See README in branch
Building Height Estimation	building_height	See README in branch

To access a module:

git checkout <branch_name>

---

📚 Documentation

Each branch contains complete documentation:

• README.md - Module-specific guide
• requirements.txt - Python dependencies
• Configuration files with inline comments

---

📄 License

MIT License. See LICENSE file in each branch.

---

Last Updated: November 6, 2025