what is ios-gaussian-splatting-demo fr?

This is an iOS demo app from Voxelio that shows how to turn a scan of a real object or scene into a 3D model, all on the phone itself, with no cloud processing involved. The technique it uses is called Gaussian Splatting, a way of representing a 3D scene as a large number of small soft blobs of color and shape rather than a traditional mesh of triangles. The app walks through a full pipeline. First, it uses Apple's ARKit to capture camera images along with the phone's position and, on supported devices, LiDAR depth information as you move around a subject. It then writes this captured data into a dataset on the device. Next, it trains a Gaussian Splat model locally using a tool called msplat, with a selectable quality target ranging from 1,000 to 5,000 points. Once training finishes, the result is saved as a resumable checkpoint and converted into a compact file format called SPZ. Finally, the app renders that SPZ file in real time using a renderer called MetalSplatter, so the user can orbit and zoom around their captured 3D object directly in the app. Everything, including the images, dataset, training checkpoints, and final result, stays in the app's local Documents folder, and the finished SPZ file can be shared or exported. The project deliberately avoids analytics, user accounts, and any cloud services. To run it, you need a physical iPhone or iPad running iOS 18 or later, since the camera capture and on-device training cannot be properly tested in the simulator. The project is set up as an Xcode project using SwiftUI. The code in this repository is source-available under the PolyForm Noncommercial License, meaning it can be viewed and used for free for noncommercial purposes, but commercial use requires a separate paid license from Voxelio. A companion finished app with the same scanning workflow is also available on the App Store.