Subset of the LaMAR dataset in the form of COLMAP models: https://lamar.ethz.ch/
The original data is available at https://cvg-data.inf.ethz.ch/lamar/

This directory contains one zip file for each of the 3 scenes CAB, HGE, and LIN.
In each zip file, you will find:
- `sparse/`: COLMAP model with camera poses, intrinsic parameters, sparse 3D point cloud.
- `images/`: All images, in JPEG format.
- `pose_covariances.txt`: CSV file with one row per image. Each row includes the image name and the flattened 6x6 covariance of the camera pose. This is available for most of images of the HoloLens and phone devices (a small number is missing) and not for NavVis. Images that do not have a covariance do not appear in this file.
- `ios_depth/`: Depth maps recorded by the LiDAR sensors of the iPad/iPhone devices, aligned with the corresponding images (names starting with `ios_`). Not all iOS images have a depth map.
- `ios_depth_confidence/`: Confidence maps associated with the iOS depth maps, as per-pixel values within three levels (high, medium, low). Guaranteed to exist if the image has a depth map.
- `navvis_depth/`: Depth maps rendered from the mesh for images captured by the NavVis scanner (names starting with `navvis_`). All NavVis images should have a corresponding depth map.

All depth maps are stored in millimeters and can be read with the following Python code: `cv2.imread(path, cv2.IMREAD_ANYDEPTH) / 1_000`.