housella-ai 1.0.1

Professional multi-task AI framework for automated 2D-to-3D floor plan recognition

Housella AI Framework

A professional multi-task AI framework for automated 2D-to-3D floor plan recognition. This framework provides advanced geometry extraction, room segmentation, and architectural feature detection from house plan images.

Features

• Multi-task Learning: Concurrent heatmap regression, room segmentation, and icon detection.
• Auto-Download Weights: Model weights are automatically fetched from cloud storage on first use to ensure a lightweight installation.
• Geometry Extraction: Converts neural predictions into structured architectural data (polygons, doors, windows).
• GPU Acceleration: Built-in support for CUDA-enabled devices.

Installation

Install via pip:

pip install housella-ai

Quick Start

from housella_ai import FloorPlanArchitect

# Initialization (automatically downloads weights if missing)
architect = FloorPlanArchitect()

# Process an image
result = architect.process_image("floorplan_sample.jpg")

# Structured output
print(f"Detected {len(result['points'])} architectural elements.")
print(f"Estimated average door width: {result['averageDoor']} pixels.")

Configuration

You can override the weights location using environment variables:

export HOUSELLA_WEIGHTS="/path/to/custom_weights.pkl"

Requirements

• Python >= 3.8
• PyTorch
• NumPy
• OpenCV
• Pillow
• Requests (for weight downloading)
• Tqdm (for progress tracking)

License

MIT License. See LICENSE for details.