README.md

metadata

license: mit
tags:
  - agent
  - collaborative
  - perception
  - drone

CoSwarm: A large-scale multi-drone collaborative perception dataset, providing a standardized benchmark for security-oriented collaborative perception research.

CoSwarm Dataset Verification & Extraction Guide

This guide covers the core steps to verify and extract the origin CoSwarm dataset.

1. Verify File Integrity (Recommended)

Ensure downloaded files are not corrupted using the checksums.md5 file:

# Navigate to the folder with downloaded files
cd ./CoSwarm

# Run integrity check
md5sum -c checksums.md5

• Success: All lines show OK (e.g., samples.tar.gz.aa: OK).
• Failure: Re-download the file marked "corrupted".

Windows Users: Replace with certutil -hashfile [filename] MD5 (e.g., certutil -hashfile samples.tar.gz.aa MD5) and compare with the value in checksums.md5.

2. Extract the Dataset

2.1 Extract Large Split Folders (samples & sweeps)

These are split into *.tar.gz.aa/*.tar.gz.ab files. Merge and extract with:

# For samples folder
cat samples.tar.gz.* | pigz -d | tar -xvf -

# For sweeps folder
cat sweeps.tar.gz.* | pigz -d | tar -xvf -

2.2 Extract Small Folders (v1.0 & maps)

These are single *.tar.gz files. Extract directly:

# For v1.0 folder
tar -xzvf v1.0.tar.gz

# For maps folder
tar -xzvf maps.tar.gz

Note: If pigz is missing, install it via sudo apt install pigz (Linux) or brew install pigz (macOS), or use gzip -d instead (slower).

3. Final Structure

After extraction, you’ll get the original folders:

CoSwarm
├── maps
│   ├── 00b9d0794bace3daa328c0273608bd62.png
│   └── 26222988e89156a507ade978f378c7e5.png
├── samples
│   ├── CAM_BOTTOM_id_0
│   ├── CAM_BOTTOM_id_1
│   ├── CAM_BOTTOM_id_2
│   ├── CAM_BOTTOM_id_3
│   ├── CAM_BOTTOM_id_4
│   ├── CAM_BOTTOM_id_5
│   ├── CAM_BOTTOM_id_6
│   ├── CAM_BOTTOM_id_7
│   ├── LIDAR_BOTTOM_id_0
│   ├── LIDAR_BOTTOM_id_1
│   ├── LIDAR_BOTTOM_id_2
│   ├── LIDAR_BOTTOM_id_3
│   ├── LIDAR_BOTTOM_id_4
│   ├── LIDAR_BOTTOM_id_5
│   ├── LIDAR_BOTTOM_id_6
│   └── LIDAR_BOTTOM_id_7
├── sweeps
│   ├── CAM_BOTTOM_id_0
│   ├── CAM_BOTTOM_id_1
│   ├── CAM_BOTTOM_id_2
│   ├── CAM_BOTTOM_id_3
│   ├── CAM_BOTTOM_id_4
│   ├── CAM_BOTTOM_id_5
│   ├── CAM_BOTTOM_id_6
│   ├── CAM_BOTTOM_id_7
│   ├── LIDAR_BOTTOM_id_0
│   ├── LIDAR_BOTTOM_id_1
│   ├── LIDAR_BOTTOM_id_2
│   ├── LIDAR_BOTTOM_id_3
│   ├── LIDAR_BOTTOM_id_4
│   ├── LIDAR_BOTTOM_id_5
│   ├── LIDAR_BOTTOM_id_6
│   └── LIDAR_BOTTOM_id_7
└── v1.0
    ├── attribute.json
    ├── calibrated_sensor.json
    ├── category.json
    ├── ego_pose.json
    ├── instance.json
    ├── log.json
    ├── map.json
    ├── sample_annotation.json
    ├── sample_data.json
    ├── sample.json
    ├── scene.json
    ├── sensor.json
    └── visibility.json

Parsed CoSwarm Dataset For Detection

If you just want to test or train the BTS, you can download and use the preprocessed dataset CoSwarm-det.tar.gz directly.

Then follow the doc of BTS to specify it.

The structure of folders as follows:

.
├── test
│   ├── agent0
│       ├── 0_0
│       ...└── 0.npy
│   ├── agent1
│   ├── agent2
│   ├── agent3
│   ├── agent4
│   ├── agent5
│   ├── agent6
│   └── agent7
├── train
│   ├── agent0
│   ├── agent1
│   ├── agent2
│   ├── agent3
│   ├── agent4
│   ├── agent5
│   ├── agent6
│   └── agent7
└── val
    ├── agent0
    ├── agent1
    ├── agent2
    ├── agent3
    ├── agent4
    ├── agent5
    ├── agent6
    └── agent7