Scanner3D

Scanner3D

Offline 3D scanning app for iPhone. Capture depth, RGB, camera pose, and confidence maps using the LiDAR sensor, then export scans and fuse them into 3D point clouds in marker frame.

Demo

https://www.loom.com/share/b4a0ab5e65ce4debafca076c697b67ae

Visualization of the fused point cloud in marker frame.

https://www.loom.com/share/dc6d9b6e6215488fb55b152fd6bb4fec

3D Gaussian Splatting reconstruction from a Scanner3D capture, viewed in the web viewer.

iOS App

Download the free iOS scanning app:

Features

• Real-time depth heatmap preview at 60fps
• Capture depth (256x192), RGB (1920x1440), intrinsics, camera pose, and confidence maps
• Save and manage multiple scan sessions in the Library
• Export scans as zip files via the iOS Files app
• ArUco marker generation for camera-to-robot calibration
• Mesh scanning mode with ARKit scene reconstruction

Requirements

• iPhone 12 Pro or newer (with LiDAR sensor)
• iOS 17.5+
• Python 3.11 (for the processing pipeline)

How It Works

1. Scan -- Point your iPhone at an object with an ArUco marker visible, tap "Start Scan", then tap "Capture" for each frame
2. Save -- Save the scan session to the Library
3. Export -- Export as a zip file and transfer to your computer via Files app
4. Process -- Run the 3-step pipeline below to get a fused point cloud in marker frame

Exported Data Format

Each scan session is exported as a zip containing capture_NNNNNN/ folders. Each capture folder contains:

File	Description
pointcloud.ply	Point cloud in camera optical frame (confidence-filtered)
depth.png	16-bit grayscale depth in millimeters
rgb.png	Color image
camera_info.json	Camera intrinsics (ROS CameraInfo format)
T_world_camera.txt	4x4 ARKit camera-to-world transform
meta.json	Capture metadata (timestamps, resolutions)
depth_vis.png	Depth visualization (red=close, blue=far)

Pipeline

3-step processing pipeline to go from raw captures to a fused point cloud in marker frame.

Setup

Requires Python 3.11 (Open3D does not support 3.13 yet).

cd pipeline
python3.11 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Run

cd pipeline
source venv/bin/activate

# Step 1: Calibrate — find ArUco marker pose from the best capture
python3 calibrate_3d.py --scan-dir ../example_scan/2026-03-23--19-45-00 --best

# Step 2: TSDF fusion — fuse all captures into a single mesh
python3 fuse_pointclouds.py --data-dir ../example_scan/2026-03-23--19-45-00 --mode tsdf --voxel-size 0.002 --icp

# Step 3: Transform — convert fused mesh from world frame to marker frame
python3 transform_pointcloud.py \
    --input ../example_scan/2026-03-23--19-45-00/fused.ply \
    --world-marker ../example_scan/2026-03-23--19-45-00/T_world_marker_3d.txt \
    --output ../example_scan/2026-03-23--19-45-00/fused_marker.ply

Output

example_scan/2026-03-23--19-45-00/
├── capture_000000/              # Raw captures (input)
├── capture_000001/
├── ...
├── T_world_marker_3d.txt        # Step 1: marker pose (from best capture)
├── fused.ply                    # Step 2: fused mesh in ARKit world frame
└── fused_marker.ply             # Step 3: fused mesh in marker frame

Pipeline Details

Step	Script	What it does
1	calibrate_3d.py --best	Detects ArUco marker in RGB, looks up LiDAR depth at corners, selects the capture with the largest marker (closest to camera = most accurate depth)
2	fuse_pointclouds.py --mode tsdf	TSDF volumetric fusion using ARKit camera poses. ICP refines pose alignment between captures. Outputs a clean triangle mesh
3	transform_pointcloud.py	Transforms the fused mesh from ARKit world frame to marker frame via inv(T_world_marker)

Key Options

Option	Description
--scan-dir / --data-dir	Directory containing capture_* folders (required)
--best	Use the single best capture for calibration (recommended)
--mode	ply, depth, or tsdf (default: ply)
--voxel-size	Voxel size in meters for TSDF (default: 0.005, use 0.002 for fine detail)
--icp	Enable ICP pose refinement before TSDF integration
--bounds	Workspace bounds filter: x_min,x_max,y_min,y_max,z_min,z_max
--output	Output PLY path