Metric scale, not arbitrary units
DepthPro provides monocular metric depth. Combined with vehicle INS or wheel odometry, the output point cloud is in meters from the first frame.
Camera reconstruction
Metric 3D from a single forward camera. No LiDAR needed.
Many field robots and early-stage AV prototypes don't carry a LiDAR. LidarFlow's camera pipeline uses Apple's DepthPro monocular metric depth model plus your INS or odometry to produce a real-world-scale, colored point cloud.
Calibrates against LiDAR when present
If your recording also has LiDAR, the camera path can calibrate against it to tighten geometry.
Moving objects masked
Pedestrians, vehicles and other dynamic objects are masked automatically so they don't bleed into the static map.
Who this is for
- · Wheeled or tracked UGVs without a LiDAR payload
- · Early-stage AV stacks recording dashcam plus GPS or INS
- · Site capture teams already running a forward camera
- · Humanoid and bipedal platforms with head-mounted cameras
What you need on the bag
- · One forward-facing camera topic
- · Vehicle odometry, IMU, or GNSS for scale and pose
- · `.mcap` or ROS1 `.bag` up to 20 GB
No LiDAR on the platform
Get metric 3D from the camera you already have.
Related: Use cases, How it works.