Skip to content

Migration to canonical scene MCAP

Raw KITTI

The former wiki/projects/kitti writer emitted generic Foxglove channels and could silently skip unsynchronized camera images. Its write_mcap(...) compatibility entry point now delegates to kitti_source_config and write_profile_mcap, producing strict lidar-camera-scene/1 output.

Preferred conversion and launch:

uv run --active python scripts/kitti_to_calibrator.py \
  --calibration /data/2011_09_26 \
  --sequence /data/2011_09_26_drive_0001_sync \
  --output output/scene.mcap

The KITTI adapter:

  • reads Velodyne and image payloads lazily;
  • derives rectified direct camera-from-LiDAR transforms from R_rect_00 and P_rect_XX;
  • converts OXTS geodetic localization to world-from-IMU poses normalized at the first frame;
  • associates synchronized OXTS packet indexes with the LiDAR timeline;
  • fails on missing files, count mismatches, invalid calibration, or synchronization gaps.

Existing Foxglove JSON/base64 MCAP

Standard Foxglove JSON recordings can be normalized explicitly:

from lidar_camera_calibrator import foxglove_source_config, write_profile_mcap

source = foxglove_source_config(
    "legacy.mcap",
    tf_topic="/tf",
    lidar_topic="/lidar/points",
    world_frame="world",
    imu_frame="imu",
)
write_profile_mcap(source, "scene.mcap")

The adapter requires the standard schema names foxglove.FrameTransform, foxglove.CameraCalibration, foxglove.CompressedImage, and foxglove.PointCloud. Camera topics pair by replacing /calibration with /image. FLOAT32 x, y, and z point fields are mandatory; intensity is optional and defaults to zero. Calibration rectification and projection offsets are folded into each canonical camera transform.

A single static world-from-IMU transform becomes a constant pose over the LiDAR timeline. Dynamic world-from-IMU transforms retain their source timestamps and must bracket the complete LiDAR range.

Direct NumPy integration

Run the complete two-camera example:

uv run --active python examples/numpy_scene.py --frames 100 --launch

Use SourceAdapterConfig directly when points and RGB arrays already exist in memory. See examples/numpy_scene.py for synthetic construction. To see real raw files loaded eagerly into plain lists—without calling the KITTI adapter—run:

uv run --active python scripts/raw_numpy_to_calibrator.py --frames 10

That script reads .bin, PNG, timestamp, calibration, and OXTS files itself, creates lists of NumPy arrays, then calls SourceAdapterConfig, write_profile_mcap, and launch_calibrator in order.

Viewer invocation

Raw paths and arbitrary MCAP files are no longer accepted by the viewer:

python -m lidar_camera_calibrator scene.mcap

Use an adapter first. This boundary keeps synchronization, transform interpretation, and malformed-data policy deterministic and testable.