README.md

---
license: cc-by-4.0
size_categories:
- 1K<n<10K
task_categories:
- object-detection
- robotics
- image-segmentation
viewer: false
tags:
- aerial-ground
- multi-robot
- person-detection
- occlusion
- ros2
- outdoor
- synchronized
---

# Co-GLANCE Dataset

[**Project Page**](https://co-glance.github.io/)

## Dataset Summary

[Co-GLANCE](https://co-glance.github.io) is a real-world aerial–ground synchronized dataset for person detection and multi-robot perception research. It provides over 4,000 synchronized RGB frames from aerial and ground viewpoints across two outdoor collection events, recorded on semi-structured terrain. Unlike simulation-based or road-scene datasets, Co-GLANCE offers raw sensor streams from heterogeneous robot platforms alongside ground-truth bounding box annotations, making it suitable for evaluating perception stacks under realistic conditions including occlusion, camouflage, and multi-person scenes.

Raw ROS 2 bag files from both platforms are also released separately to support broader evaluation of perception and autonomy stacks beyond static image benchmarks.

## Scenarios

The dataset is organized into two collection scenarios:

| | Construction Scenario (2026-03-30) | Camouflage Scenario  (2026-04-14) |
|---|---|---|
| **Scenario** | A construction worker walks through a construction site, followed by a ground robot and an aerial robot recording the scene. | Two individuals wearing camouflage attempt to move through a visually occluded area. |
| **Ground Hardware** | GoPro HERO 10 | Boston Dynamics Spot — front-left and front-right cameras (stitched) |
| **Aerial Hardware** | Arducam HQ IMX477 | GoPro HERO 10 |
| **No. of Runs** | 4 | 3 |

## Preview
![Alt text](construction_preview.png)
![Alt text](camouflage_preview.png)

## Data Categories

Each run is split into scene-type categories depending on the event:

| Category | Construction Scenario (03-30) | Camouflage Scenario (04-14) |
|---|---|---|
| **Clean** | Scenes with exactly one person clearly visible. Represents the primary target for single-person detection and tracking. | Scenes where both individuals are at least partially visible. |
| **Filtered Out** | Scenes with no humans present. Excluded from primary detection analysis. | — |
| **Multiperson** | Scenes where more than one person is visible simultaneously. | — |
| **Partial Occlusion** | — | Scenes where one individual is completely occluded by environmental features. |
| **Full Occlusion** | — | Scenes where both individuals are completely occluded, in at least one viewpoint. |

## Frame Counts

Per-run frame counts for each scene category. The dataset contains **2,071 annotated frame pairs** in total.

### Construction Scenario (03-30)

| Run | Clean | Filtered out | Multiperson | All |
|---|---|---|---|---|
| 1 | 77 | 34 | 7 | 118 |
| 2 | 229 | 43 | 54 | 326 |
| 3 | 88 | 54 | 138 | 280 |
| 4 | 135 | 144 | 206 | 485 |
| **Total** | **529** | **275** | **405** | **1209** |

### Camouflage Scenario (04-14)

| Run | Clean | Partial Occlusion | Full Occlusion | All |
|---|---|---|---|---|
| 1 | 54 | 101 | 31 | 186 |
| 2 | 210 | 252 | 83 | 545 |
| 3 | 69 | 47 | 15 | 131 |
| **Total** | **333** | **400** | **129** | **862** |

## Dataset Structure

```
dataset/
├── 03-30/                        # Construction Scenario
│   └── <run>/                    # 4 runs
│       ├── all/
│       ├── clean/
│       ├── filtered_out/
│       └── multiperson/
│           ├── aerial/
│           │   ├── rgb/          # PNG frames
│           │   ├── gt_bbox/      # Frames with bounding boxes rendered on image
│           │   ├── acceleration/ # JSON sensor stream
│           │   ├── attitude/     # JSON sensor stream
│           │   ├── gps_position/ # JSON RTK GPS stream
│           │   └── *.json        # Additional odometry data
│           └── ground/
│               └── <same structure>
└── 04-14/                        # Camouflage Scenario
    └── <run>/                    # 3 runs
        ├── all/
        ├── clean/
        ├── partial_occlusion/
        └── full_occlusion/
            ├── aerial/
            │   └── <same structure>
            └── ground/
                └── <same structure>
```

## Data Fields

Each aerial and ground platform folder contains the following topics:

| Folder / File | Format | Description |
|---|---|---|
| `rgb/` | PNG | Raw RGB frames from the platform camera |
| `gt_bbox/` | PNG | Corresponding frames with ground-truth bounding boxes rendered on the image |
| `acceleration/` | JSON | IMU linear acceleration stream |
| `attitude/` | JSON | Platform orientation / attitude stream |
| `gps_position/` | JSON | RTK GPS position stream |
| `*.json` | JSON | Additional odometry and platform state data |


## Citation

```bibtex
@article{co-glance2026,
  title={Co-GLANCE: Uncertainty-Aware Active Perception for Heterogeneous Robot Teaming},
  author={Redacted until publication},
  journal={Redacted until publication},
  year={2026}
}
```

## License

This dataset is released under the [Creative Commons Attribution 4.0 International License (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/).