B-COD Dataset

This repository contains an anonymized subset of a marine navigation dataset, providing multimodal sensor data for the B-COD project.

Please download the dataset from this Google Drive link.

Original Dataset Structure

Each data snippet contains:

• Belief raster (64×64×5 float16)
• Map slice (64×64×3 semantic image)
• Goal mask (64×64 binary)
• Sensor flags (5-bit uint8 vector)
• Ground truth trajectory (∆x,∆y,∆ψ for 8 waypoints)
• Waypoint log-variances
• Metadata (latitude/longitude, weather, clip ID)

Dataset Statistics

• Modalities: 100% contain LiDAR and IMU data

  • Day camera: 72%
  • Night camera: 28%
  • GNSS: 64%
  • Sonde: 18%

• Belief spread:

  • Median planar 1σ = 0.38m
  • 95th percentile = 2.1m

• Lighting conditions: 0.2-55 kLux

• Obstacle ranges:

  • Mean: 14.2m
  • Minimum: 0.8m

Current Publicly Available Dataset Directory Structure (Please note that this is a subset of the original dataset to maintain anonymity of the review process)

.
├── dataset_20250226_150909/
│   └── zEeZKgHmazs1R6C.h5
├── dataset_20250228_063657/
│   └── sA0JsWGpDDFMUMK.h5
...
└── dataset_lidar_json_untagged/
    └── *.json

Download

To download the dataset, please follow this this Google Drive link.

Usage

# PyTorch
from src.pytorch_loader import DatasetLoader
dataset = DatasetLoader("path/to/dataset")

# JAX
from src.jax_loader import DatasetLoader
loader = DatasetLoader("path/to/dataset", batch_size=32)
batch = loader.get_batch(0)

License

License will be released after the review process. We plan to release the whole dataset (280GB) under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).