🏠 Floor Plan Wall Detection using DeepLabV3+ 🧱

📝 Overview

This project implements a deep learning model for automatic wall detection in architectural floor plans using the DeepLabV3+ architecture. The model is trained on the zimhe/pseudo-floor-plan-12k dataset from HuggingFace.

🌟 Key Features

• Multiple Detection Methods: UNet, DeepLabV3+, and direct image processing
• Wall Coordinate Extraction: Converts wall masks to line coordinates
• 3D Visualization Ready: Outputs coordinates in JSON format for 3D rendering

🔄 Model Comparison

Model	Advantages	Use Case
UNet (UnetModel.ipynb)	🚀 Faster training 🎯 Better accuracy 🧠 Lower memory usage	Recommended for all wall detection tasks
DeepLabV3+ (main.py)	🔍 Higher capacity 🌐 Pre-trained backbone	Complex floor plans with many details
Direct Processing	⚡ No training required 📊 Works on binary images	When you already have wall masks

🛠️ Installation

Step 1: Clone the repository

git clone https://github.com/Divak-ar/floorData.git
cd floorData

Step 2: Create a virtual environment

# For Windows
python -m venv .venv
.venv\Scripts\activate

# For macOS/Linux
python3 -m venv .venv
source .venv/bin/activate

Step 3: Install dependencies

pip install -r requirements.txt

📊 Dataset

The project uses the pseudo-floor-plan-12k dataset from HuggingFace. To download the dataset:

python dataDownload.py

This will:

• Download 100 floor plan images and their corresponding wall masks
• Save them in the huggingface_dataset directory

🚀 Usage

Training the model

python main.py

The training process:

1. Loads and preprocesses floor plan images and wall masks
2. Splits data into training and validation sets
3. Trains a DeepLabV3+ model with ResNet50 backbone
4. Saves the best model during training
5. Visualizes prediction results

Using UNet Model (Recommended)

Open and run the UnetModel.ipynb notebook for better wall detection results.

Inference on new images

from main import predict_on_new_image
from tensorflow.keras.models import load_model

# Load trained model
model = load_model('huggingface_dataset/model/deeplab_walls_best.h5')

🔧 Model Architecture

The wall detection model uses a DeepLabV3+ architecture with:

• ResNet50 backbone pre-trained on ImageNet
• Atrous Spatial Pyramid Pooling (ASPP) module
• Decoder with skip connections from encoder
• Binary segmentation output (walls vs. background)

📈 Results

The UNet model typically achieves:

• Higher IoU (Intersection over Union) for wall segmentation
• Better line detection accuracy
• Faster inference time

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.