3d-reconstruction-toolkit

3d-reconstruction-toolkit lets you easily reconstruct 3D models based on video sequence using provided application and your own hardware. Reconstruct 3D models via classic COLMAP-OpenMVS pipeline or its more modern variant (ALIKED + LightGlue) choosing reconstruction options based on your preferences and hardware limitations.

Contents

• Why?
• Installation
• Usage

Why?

I wanted a tool that allows you to:

• Reconstruct 3D models with barely any setup.
• Reconstruct 3D models based on your preferences (quality, computation time, hardware limitations).
• Reconstruct 3D models on the go - record video, choose options, reconstruct!
• Share 3D models of real items.
• View reconstructed 3D models without any additional tools.
• Use beautifull minimalistic interface.

3d-reconstruction-toolkit checks all of those boxes.

Installation

---

Install this repository and setup conda environment

$ git clone https://github.com/MarAn13/3d-reconstruction-toolkit.git && cd 3d-reconstruction-toolkit
$ conda env create --name <ENV_NAME> --file environment.yaml

Warning Don't forget to install dependencies first!

Change 2 lines in the app_pipeline.sh file

colmap_dir="<PATH TO COLMAP BINARY>"
openmvs_dir="<PATH TO OPENMVS BINARY>"

Place vocab_tree_file in the root of the project (cloned repository).

Dependencies

---

• COLMAP-3.9.1
• OpenMVS-2.3.0
• LightGlue

Download COLMAP and OpenMVS binaries.

Download vocab_tree_file.

Clone and install LightGlue

$ git clone https://github.com/cvg/LightGlue.git && cd LightGlue
$ python -m pip install -e .

Usage

---

Note Setup ssh server on your local machine first (e.g. setup OpenSSH on Windows 11).

1. Open the app.
2. Provide ssh parameters from the settings page.
3. Record video.
4. Select reconstruction options.
5. Open history page.
6. Enjoy!

Reconstruction examples

---

Some resulting 3d models and its statistics available via google drive.