scier/MetalSplatter

Render Gaussian Splats using Metal on Apple platforms (iOS/iPhone/iPad, macOS, and visionOS)

What it solves

MetalSplatter is a library designed to render 3D Gaussian Splats on Apple platforms. It allows developers to visualize 3D scenes captured via Gaussian Splatting techniques, enabling real-time radiance field rendering on iOS, macOS, and visionOS (including stereo rendering for Vision Pro).

How it works

The library provides a core rendering engine based on Metal, Apple's graphics API. It includes several modules to handle the data pipeline:

  • MetalSplatter: The core rendering library.
  • PLYIO: A standalone module for reading and writing binary or ASCII PLY files.
  • SplatIO: A layer that interprets PLY and .splat files as sets of 3D splats.
  • SampleApp: A minimal demonstration app to show how to integrate the library.

Who it’s for

This project is for developers building apps for Apple platforms who want to integrate 3D Gaussian Splatting visualization into their own applications, particularly those targeting spatial computing on Vision Pro.

Highlights

  • Apple Ecosystem Support: Native support for iOS, macOS, and visionOS.
  • Multiple File Format Support: Compatible with PLY, SPZ, and .splat files.
  • Stereo Rendering: Specifically optimized for stereo rendering on Vision Pro using amplification.
  • Modular Design: Includes separate IO modules for handling 3D scene data.

Related

  • Project
    bhimrazy/receipt-ocr

    An OCR engine that extracts either raw text or structured JSON data from receipt images using Tesseract and LLMs.

  • Project
    basler/pypylon

    Official Python bindings for Basler pylon C++ APIs, enabling Python applications to control Basler machine vision cameras and perform image processing tasks.

  • Project
    zhiqwang/yolort

    A runtime stack for YOLOv5 that simplifies object detection deployment by embedding pre- and post-processing into the model graph for various hardware accelerators.

  • Project
    Tencent/YOLO-Master

    A YOLO-style framework for real-time object detection that uses Mixture-of-Experts (MoE) to adaptively allocate computation based on scene complexity for higher precision and lower latency.