A production-grade, geometry-only pipeline for 3D indoor scene semantic segmentation. This project segments raw point clouds into structural elements (floor, walls, ceiling) and furniture objects using unsupervised clustering and rule-based heuristics — zero deep learning required.
For a deep-dive into the project's logic and architecture, please refer to the following:
- Technical Pipeline Guide (PIPELINE.md): Installation, usage, and geometric heuristics.
- Implementation Roadmap (implementation.md): Full architectural breakdown and research references.
| Final Semantic Segmentation | 2D Top-Down Projection h |
|---|---|
 |
 |
| Color-coded: Floor (Brown), Walls (Blue), Furniture (Green) | Occupancy grid with identified furniture footprints |
- Structural Segmentation: Iterative RANSAC with normal-alignment checks for floor, ceiling, and walls.
- Object Clustering: DBSCAN-based clustering for furniture and clutter.
- Semantic Heuristics: Automatic labeling based on Z-distribution, surface normals, and aspect-ratio analysis.
- Bounding Boxes: Axis-Aligned (AABB) and Oriented (OBB) bounding box estimation with dimensions.
- Interactive Viewer: Custom GUI for real-time manual inspection and labeling.
- 2D Mapping: Automated generation of top-down occupancy and semantic maps.
- Robust Testing: 143+ unit and integration tests covering the entire pipeline.
The pipeline follows a two-stage expert system approach to ensure clean separation between structural surfaces and furniture:
- Stage A (Structural): Uses RANSAC to "peel" away high-density planar primitives (floor, walls, ceiling).
- Stage B (Object): Performs DBSCAN Euclidean clustering on the non-planar residuals to isolate individual furniture units.
- Stage C (Refinement): Computes geometric properties (centroids, dimensions, orientation) for the final semantic report.
# Clone the repository
git clone https://github.com/ERICR-recruiter/l02-DsThakurRawat.git
cd l02-DsThakurRawat
# Setup environment
python3 -m venv venv
source venv/bin/activate
# Install dependencies
pip install -r requirements.txtPerfect for immediate testing:
python3 scripts/generate_synthetic_room.py --output data/synthetic/room_01.plypython3 main.py --input data/synthetic/room_01.ply --config configs/default.yamlpython3 -m src.interactive_viewer --input outputs/segmented_room.plyEvery run produces a standard artifacts bundle in the outputs/ directory:
segmented_room.ply: Fully labeled 3D point cloud.segmentation_report.json: Pydantic-validated JSON containing IDs, dimensions, and labels for all clusters.segmentation_viz.png: High-resolution 2D semantic map.
The project maintains a rigorous test suite with 143 tests covering all geometric heuristics.
python3 -m pytest tests/ -v- Name: Divyansh Rawat
- Contact number(s): +91 6261283255
- Email address(es): divyanshthakur594@gmail.com
- GitHub Username: DsThakurRawat
Good luck! This assignment will test your ability to reason about geometry and clustering in 3D scenes.