moonbit-robot-vision-formats

Offline format bridges for robotics vision datasets in MoonBit.

The project focuses on text artifacts commonly exported from ROS bag workflows: image timestamp indexes, CameraInfo-style calibration YAML, 2D trajectory CSV, depth image metadata and object annotations. It is not a ROS client and does not subscribe to topics. The goal is to give MoonBit users a small, testable library for converting and aligning perception data before it reaches SLAM, calibration or dataset review tools.

///|
test "README quick start" {
  let csv =
    #|sec,nsec,frame_id,path
    #|42,0,camera/front,front_000001.png
    #|42,33333333,camera/front,front_000002.png
    #|
  let frames = parse_image_index_csv(csv)
  inspect(
    image_index_summary(frames),
    content="2 frames from 42.000000000 to 42.033333333",
  )
}

Scope

• Parse and write image timestamp CSV exported from bag extraction scripts.
• Parse the scalar and matrix subset of camera calibration YAML used by common sensor_msgs/CameraInfo dumps.
• Parse trajectory CSV containing timestamped planar poses.
• Parse depth metadata sidecars without loading large depth image payloads.
• Parse 2D annotation CSV and align all streams by timestamp tolerance.

Repository Notes

This repository was created as an original MoonBit OSC2026 ecosystem library. Before implementation, the package space on Mooncakes was checked with robotics, vision, camera, depth, CSV and YAML keywords. No mature MoonBit package with this same offline robotics vision bridge scope was found.

Validation

Run these commands before submitting or publishing:

moon check --target all
moon test --target all
moon check --target all --deny-warn --fmt
moon fmt --check
moon info
git diff --exit-code

License

Apache-2.0.