Gaussian SplatKing vs AR Recorder for Datasets

Gaussian SplatKing is a capture app for Gaussian splatting and radiance field reconstruction. It produces reconstruction-ready input data for any downstream pipeline and, for LiDAR captures, refines camera poses and trains a Gaussian splat on the device. Capture • Photo and video fire the 0.5x ultra-wide and 1x wide lenses simultaneously. A single-lens (1x) option, with RAW (DNG) photo capture, is available in Settings. • LiDAR mode pairs ARKit depth with the 1x wide lens, writes native COLMAP files (cameras, images, sparse points) on-device, and shows the point cloud as it accumulates. • 4K/30 HEVC video with Standard and Max Fidelity modes; optional undistorted video exports. • Auto photo cadence for walk-through capture, with a selectable interval. • Manual controls applied to both lenses: ISO, shutter speed or shutter angle, white balance, exposure, AF Lock, manual focus distance, tap-to-focus; a minimum shutter speed. • Optional capture location metadata, off by default. • iPhone and iPad. Process • Pose refinement: after a LiDAR capture, camera poses are refined on-device (feature matching and bundle adjustment) and recorded in the export. Free for everyone. • Train Splat (SplatKing Pro): train a Gaussian splat from a refined LiDAR capture on the device. No upload, no account. Output is an .spz file with an in-app viewer. Export • Every capture is a zip with media, metadata, COLMAP files where applicable, and splatpack.json, shared through the Files app or the share sheet to any pipeline. • Session quality summaries and per-file quality flags. SplatKing Pro is a one-time purchase that adds focus peaking, an RGB histogram, capture presets, the capture-interval selector, and on-device training. Everything that is free today stays free, and Pro never changes free capture quality. Analytics are anonymous and, in regions that require consent, off unless you opt in. Capture content never leaves the device unless you share it.

■ Key Features • Dataset Recording Records camera frames, LiDAR data, camera poses, and GPS coordinates, synchronized into a single video file. • Native & Web Scan Modes Alongside the high-accuracy ARKit-based native mode, a web mode records datasets for use in web environments — capturing camera video synchronized with motion, orientation, and GPS data. Switch modes right from the camera screen. • Real-Time Mesh Visualization Overlays the spatial mesh detected by the LiDAR sensor on screen, so you can check scan coverage at a glance. • Real-Time Camera Path Visualization Overlays the camera's trajectory on screen while recording, making the scan path easy to follow. • Frame Info Overlay View camera intrinsic and extrinsic parameters, pose data, and AR tracking state in real time. • Multiple Video Formats Choose the resolution and frame rate that fit your needs. • Dataset Decoding & Sharing Decodes videos into per-frame data (RGB images, camera trajectory and parameters, depth, and more), exports them as an archive, and shares it. • Save to Photos Turn on the option to save recordings straight to the Photos app. • Library Play, inspect, share, and delete recordings in one place, with native and web recordings organized separately. ※ LiDAR data capture and mesh visualization are available only in native mode on devices equipped with a LiDAR sensor. ■ Use Cases • Building AR / VR content datasets • Computer vision, SLAM, and 3D reconstruction research • Collecting machine learning training data • Developing and validating web-based AR / vision features • Spatial computing prototyping • Indoor and outdoor environment scanning ■ Privacy • Captured datasets are stored only on your device and are never transmitted externally. • Deleting the app immediately removes all data from your device. • No sign-up or login required.