elevate3d 0.5.1

3D terrain and structure reconstruction from single RGB images

Elevate3D

Generate 3D models from satellite & aerial images using deep learning.

---

Disclaimer

Elevate3D is an experimental, early-stage project.
It's not production-ready, and results may be inconsistent depending on the input. Expect rough outputs, strange artifacts, and occasional surprises.

---

Features

• Automatic Building Segmentation (Mask R-CNN)
• Elevation Prediction from RGB (Pix2Pix)
• 3D Mesh Generation (Open3D)
• Tree Detection with DeepForest
• Pretrained Models - No training required
• End-to-End Pipeline - From image to interactive 3D output

---

Installation

Install with pip:

pip install elevate3d

---

Usage

1. Web Interface (Recommended)

Run the local web app:

elevate3d-run

This launches a local server where you can upload images and view results interactively in your browser.

2. Python API

You can also run the pipeline programmatically:

from elevate3d.run_pipeline import run_pipeline

run_pipeline("path_to_your_image.jpg")

This processes the image and opens a viewer window showing the 3D model.

---

Input Requirements

• Accepts aerial or satellite RGB images (.jpg, .jpeg, .png)
• Images should ideally be top-down and contain visible buildings or tree cover

---

How It Works

• Mask R-CNN segments buildings
• Pix2Pix generates DSM from RGB
• Tree detection adds tree geometry
• ResNet-50 roof type detection
• Open3D constructs the final mesh (as .glb)

---

License

MIT License. See LICENSE for details.