RoboCup HSL Object Detection Images with Bounding Boxes

HSL Objects v1 contains 5,037 images annotated for object detection in the RoboCup Humanoid Soccer League. The dataset is split into 4,029 training, 504 validation, and 504 test images. It was created by whirlwind-ams and last updated in April 2026.

Use Cases

• Train object detection models to locate 'Ball' and 'Goalpost' bounding boxes in soccer field images.
• Evaluate model performance on specific field markings like 'Penalty Marks', 'L-Intersections', 'T-Intersections', and 'X-Intersections'.
• Benchmark detection algorithms on the 'K1' class using the provided validation and test splits.
• Develop vision systems for robots by leveraging COCO-style bounding box annotations across 7 object classes.

Strengths

• 5,037 annotated images provide a substantial base for training.
• Predefined splits of 4,029 train, 504 validation, and 504 test images facilitate standard evaluation.

Limitations

• Limited to 7 object classes specific to RoboCup HSL, reducing generalizability.
• No information on image resolution, diversity of scenes, or potential label noise.

Provenance

Source

whirlwind-ams

Collection Method

Images annotated with COCO-style bounding boxes for RoboCup HSL.

Freshness

Last updated April 2026.