ooolabdev/ooosplat

A local desktop application that converts videos into 3D Gaussian Splatting.

What it solves

OOOSplat is a desktop application that simplifies the process of converting local videos into 3D Gaussian Splatting (3DGS) models. It removes the need for users to manually set up complex command-line tools or remote services by bundling all necessary engines into a single local workflow.

How it works

The application uses a Rust-based backend and a React frontend to orchestrate a pipeline of specialized tools:

  1. Video Analysis & Extraction: Uses FFprobe to analyze the video and FFmpeg to extract frames based on selected quality settings (Fast, Balanced, or Fine).
  2. Structure from Motion: Employs COLMAP to perform feature extraction, matching, and camera reconstruction. It automatically detects if an NVIDIA GPU is available for CUDA acceleration, falling back to CPU if necessary.
  3. Training: Uses Brush to train the Gaussian Splats based on the reconstructed cameras and images.
  4. Export: Validates and publishes the final result as a .ply file.

Who it’s for

It is designed for users who want to create 3D Gaussian Splatting models from video on Windows, macOS (Apple Silicon), and Ubuntu 24.04 without having to manually configure separate AI and computer vision libraries.

Highlights

  • All-in-one Bundling: Includes built-in versions of FFmpeg, FFprobe, COLMAP, and Brush.
  • Cross-Platform Support: Native builds for Windows, Apple Silicon macOS, and Ubuntu 24.04.
  • Automated Pipeline: Handles everything from frame extraction to final PLY generation with a few clicks.
  • Hardware Auto-Detection: Automatically switches between CUDA GPU and CPU for COLMAP based on system capabilities.
  • Quality Presets: Offers three quality tiers (Fast, Balanced, Fine) that adjust frame retention and training iterations.

Related

  • Project
  • Project
  • Project
  • Project
  • Dispatch