nerako/insight9-calibration-20260923

Looper Insight 9 Calibration Capture

Two calibration-oriented recordings captured with a Looper Insight 9 on 2026-09-23. The release contains native-orientation stereo grayscale images, RGB JPEG images, device timestamps, raw IMU measurements, auxiliary device-estimated poses, capture metadata, quality reports, review videos, and validated ROS1 bags.

This is a raw sensor-data release, not a calibrated benchmark or ground-truth trajectory dataset. No intrinsics, extrinsics, or camera–IMU solution has been computed from these recordings. Target specifications and calibration suitability remain to be confirmed.

Sessions

There are 12,356 source image files across the two sessions. Both sessions were recorded under the camera_intrinsics UI label. IMU was recorded simultaneously; the label and presence of IMU do not establish adequate motion excitation for camera–IMU calibration. Short trials, static tests and debugging sessions are excluded from this release.

Contents and formats

datasets/
  20260923_185932_927943_camera_intrinsics/   # A
  20260923_190311_050131_camera_intrinsics/   # B
    left/data/*.pgm, left/data.csv
    right/data/*.pgm, right/data.csv
    rgb/data/*.jpg, rgb/data.csv
    imu.csv, vio.csv
    stereo_metadata.bin, imu_reports.bin, vio_reports.bin
    session.json, quality_report.json
rosbags/            # A/B: three camera topics + IMU
previews/           # downscaled 10 fps review videos, posters and source-frame mappings
tools/
  read_dataset.py   # read-only Python API and CLI
  export_kalibr.py  # optional ROS1 export tool
  catalog.json
  checks/          # full audits, timestamp gaps and source SHA-256
config/            # pending target/IMU configuration templates
docs/USAGE.md
session_inventory.csv
SHA256SUMS.csv

All original image geometry and orientation are preserved. No source image was rotated, cropped, rectified, or interpolated. The RGB transport's custom timestamp APP1 segment was removed from the stored JPEG; the compressed image content was not re-encoded, and the timestamp is retained in CSV. Full RGB USB packets were not retained separately.

The device reported firmware/software 2.1.0 and calibration mode enabled. Depth was disabled. This release does not include depth, disparity, point clouds, a SLAM map, audio, per-frame exposure/gain, or a new calibration solution.

Timing and coordinate conventions

Image-index columns are sequence, device_timestamp_ns, original_timestamp_ns, host_unix_ns, host_monotonic_ns, filename, file_bytes. filename is relative to its camera folder, such as left/.

• —Read timestamps as integers. Device timestamps are expressed in nanoseconds based on observed increments; they are not Unix timestamps. Subtract a common integer origin before converting to relative seconds.
• —Pair left/right images by sequence, corresponding to the same USB stereo packet. In these captures, the original right timestamp is zero. The effective right device_timestamp_ns uses the paired left timestamp while original_timestamp_ns preserves zero. This is packet association, not an independent measurement of exact exposure simultaneity.
• —RGB and stereo have different rates. Do not associate them by row number. Use device timestamps and retain the association offset.
• —Host arrival timestamps are retained for diagnostics, not substituted for device timestamps. Treat sessions separately; do not assume a continuous device clock across resets or mode changes.
• —IMU values and axis order are preserved. Stationary acceleration magnitude around 9.8 supports an m/s² interpretation. The bag exports angular velocity as rad/s, but controlled-angle verification and physical axis conventions remain pending.
• —VIO is unmodified device-estimated output. Its coordinate frame, initialization/reset behavior and absolute accuracy have not been independently verified; it is not ground truth.

Quality checks and known issues

All 12,356 source images decoded successfully. Every retained IMU/VIO binary report and stereo metadata timestamp was checked against its CSV representation. Source file sizes, checksums, timestamp monotonicity, rates, gaps and stereo pairing were checked. ROS bags were fully read back to verify topic counts and header timestamps; sample messages were also checked against source pixels and IMU values.

The strict continuity check does not pass for either session:

Normal stereo intervals are approximately 50 ms. No missing frames or timestamps were synthesized to hide these gaps. Precise locations are in `tools/checks/timestamp_gaps.csv`. session.json status complete means saving completed; it does not mean gap-free or calibration-ready.

The target was described by the contributor as AprilGrid, but target family/layout, IDs, dimensions and actual printed scale have not been independently verified. Target detection rate, spatial coverage, blur and motion excitation have not been evaluated. Pending configuration files contain null values and must not be used directly for solving. IMU noise and bias random-walk parameters are also pending.

ROS1 bags

• —`rosbags/A_all_cameras_imu.bag`
• —`rosbags/B_all_cameras_imu.bag`

The bags use lossless LZ4 container compression and preserve device timestamps in both message time and header.stamp.

Read original JPEGs if color is needed. IMU orientation is unavailable (orientation_covariance[0] = -1); the other zero covariance fields are unspecified, not measured zero noise. VIO and raw binary reports are retained in the source folders but are not included in these calibration bags.

Download and read

This repository is a file-based sensor archive, not a tabular Hugging Face datasets split. Use snapshot_download, hf download, or download a bag directly.

from huggingface_hub import snapshot_download

root = snapshot_download(
    repo_id="nerako/insight9-calibration-20260923",
    repo_type="dataset",
    local_dir="insight9-calibration-20260923",
    allow_patterns=["datasets/**", "tools/**", "config/**", "docs/**", "README.md", "SHA256SUMS.csv"],
)

To download only ready-made ROS bags, use allow_patterns=["rosbags/**", "README.md", "docs/**"] instead. The two bags together are about 8.57 GB; raw-session files total about 3.25 GB. Downloading both duplicates representations of the same image/IMU data.

After downloading the source folders and tools:

python -m pip install Pillow numpy
python tools/read_dataset.py list
python tools/read_dataset.py sample A --camera left --frame 100
python tools/read_dataset.py verify

See `docs/USAGE.md` for the Python interface, timing rules and ROS usage. Review MP4/WebM videos are downscaled, sampled at 10 fps and lossy; use source images or bags for calibration. Posters show the fifth second. Review videos are not synchronized-exposure evidence.

License and provenance

No reuse license has been specified by the contributor. Contact the repository owner about reuse terms. This dataset was captured locally from a Looper Insight 9; it is not an official Looper release. The quality checks characterize file/transport integrity, not calibration accuracy.