Gaussian Splat Localization

SplatOverflow

Research Project - CHI 2025

Real-time localization in Gaussian splat scenes and SfM models

Key Features

  • Python library for real-time localization in Gaussian splat scenes
  • Optimized PyTorch pipeline for faster localization
  • Windows-compatible Gaussian splatting software
  • WebXR interface for collaborative hardware projects
View Paper

Project Details

Real-time Localization

Developed a Python library that enables real-time localization in pre-made Gaussian splat scenes and Structure from Motion (SfM) models.

Performance Optimization

Optimized the PyTorch pipeline by pre-loading models and optimizing structure for smaller-scale scenes, significantly reducing localization time.

Cross-Platform Support

Converted open-source Gaussian splatting software to be Windows-compatible using CUDA and COLMAP documentation.