Floor Plan to 3D Pipeline

This project is a comprehensive solution for converting 2D floor plan images into interactive 3D models. It utilizes deep learning for semantic understanding and morphological image processing for 3D geometry generation.

Pipeline Visualization

The following images demonstrate the complete workflow from floor plan input to 3D model output:

Original Floor Plan Input floor plan with labeled offices and rooms	Semantic Segmentation Output Colored segmentation map showing detected rooms and structural elements
Overlay Visualization Floor plan with color-coded segmentation overlay showing pipeline processing	3D Extruded Model Final 3D model with extruded walls and room structure

Project Structure

The project is organized into three main modules:

1. SplitFloor: A deep learning module responsible for detecting structural elements (Walls, Doors, Windows).
2. SemaFloor: A deep learning module dedicated to detecting room boundaries and habitable areas.
3. ForgeFloor: The core pipeline that orchestrates the neural networks, processes the geometric data, and extrudes the final 3D model.

Installation

Ensure you have Python 3.8+ installed. Install the required dependencies:

pip install torch torchvision numpy matplotlib opencv-python pillow pyvista scikit-image tqdm pandas albumentations scipy scikit-learn

Note: A CUDA-capable GPU is recommended for training and fast inference.

Usage

The main entry point for the pipeline is located in the ForgeFloor directory.

Graphical User Interface (GUI)

To launch the interactive application with live preview:

python ForgeFloor/extruder.py --gui

Command Line Interface (CLI)

To process an image automatically:

python ForgeFloor/extruder.py --input path/to/floorplan.jpg --auto --output_dir outputs/

Workflow

1. Input: A raw 2D floor plan image (JPG/PNG).
2. Inference:
   • SplitFloor extracts structural masks (walls, windows, doors).
   • SemaFloor extracts room masks.
3. Processing: ForgeFloor cleans the masks using morphological operations.
4. Extrusion: The cleaned 2D data is extruded into 3D meshes.
5. Output: A 3D OBJ model, visualization overlays, and segmentation CSVs.