README.md

---
dataset_info:
  features:
  - name: image_id
    dtype: int64
  - name: image
    dtype: image
  - name: width
    dtype: int64
  - name: height
    dtype: int64
  - name: objects
    struct:
    - name: area
      list: float32
    - name: bbox
      list:
        list: float32
        length: 4
    - name: category
      list:
        class_label:
          names:
            '0': Ball
            '1': Goalpost
            '2': K1
            '3': L-Intersections
            '4': Penalty Marks
            '5': T-Intersections
            '6': X-Intersections
    - name: id
      list: int64
  splits:
  - name: train
    num_bytes: 4072008285
    num_examples: 4029
  - name: validation
    num_bytes: 507232886
    num_examples: 504
  - name: test
    num_bytes: 517076536
    num_examples: 504
  download_size: 5096513815
  dataset_size: 5096317707
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
task_categories:
- object-detection
tags:
- robocup
- robotics
- soccer
size_categories:
- 1K<n<10K
license: mit
---

# HSL Objects v1

5,037 images with COCO-style bounding box annotations for object detection in the RoboCup Humanoid Soccer League (HSL). Split 80/10/10 into train (4,029), validation (504), and test (504).

## Classes

| ID | Class           |
|----|-----------------|
| 1  | Ball            |
| 2  | Goalpost        |
| 3  | K1              |
| 4  | L-Intersections |
| 5  | Penalty Marks   |
| 6  | T-Intersections |
| 7  | X-Intersections |

## Sample Structure

Each example contains:

- `image_id` (int64) -- unique image identifier
- `image` (Image) -- RGB image
- `width` / `height` (int64) -- image dimensions in pixels
- `objects` -- annotation dictionary:
  - `id` (list[int64]) -- object instance IDs
  - `category` (list[int64]) -- category IDs (1--7)
  - `bbox` (list[list[float64]]) -- bounding boxes in COCO format `[x, y, width, height]`
  - `area` (list[float64]) -- bounding box areas in pixels

## Sources

- RoboCup Asia Pacific Beijing Masters 2025
- RoboCup Salvador 2025
- RoboCup German Open 2025
- RoboCup German Open 2026
- Intelligent Robotics Lab, University of Amsterdam

## Usage

```python
from datasets import load_dataset
ds = load_dataset("whirlwind-ams/hsl-objects-v1")
```

## Citation

If you use this dataset, please cite:

```bibtex
@misc{hsl_objects_v1,
  title={HSL Objects v1},
  author={Whirlwind Amsterdam},
  year={2025},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/whirlwind-ams/hsl-objects-v1}}
}
```