This project was developed as part of an internship at the University of Trento and focuses on 3D reconstruction from synthetic datasets using two main techniques: COLMAP and SuGaR.
- Project Overview
- Dataset Preparation
- Methods Used
- Results and Analysis
- Dataset Information
- Visualization of Results
The goal of the project is to perform 3D reconstruction from synthetic datasets using two methods:
- COLMAP: A tool that uses Structure-from-Motion (SfM) and Multi-View Stereo (MVS) to generate 3D models.
- SuGaR: A newer method based on 3D Gaussian Splatting.
The project evaluates these two methods in terms of performance, accuracy, and the quality of the resulting models.
Nine synthetic datasets were generated using the nerf-dataset-creator-plugin. The datasets include both outdoor (5) and indoor (4) scenes.
- COLMAP is a photogrammetry software used to perform 3D reconstruction. It combines Structure-from-Motion (SfM) with Multi-View Stereo (MVS) to create accurate point clouds and mesh models.
- SuGaR is a newer approach for 3D reconstruction that utilizes 3D Gaussian Splatting. While visually compelling, SuGaR is evaluated in this project for its performance compared to the traditional COLMAP method.
The project compares COLMAP and SuGaR in terms of:
- Accuracy of the resulting 3D models
- Execution speed
- Handling of flat and low-detail surfaces
The comparison is performed using CloudCompare for visual and quantitative evaluation of the generated models.
| Info | Value |
|---|---|
| Number of datasets | 9 |
| Dataset Type | Indoor/Outdoor |
| File Size | ~1GB per model |
| Images per dataset | 250 |
| Format | PNG |
| Resolution | 800x800 |
You can view the 3D reconstruction results on Google Drive at the following link: