Tennis Player Actions Dataset

Structured visual dataset for the recognition of sports actions, illustrating 4 key tennis gestures (serve, backhand, forehand, waiting position) with COCO annotations and body keypoints according to the OpenPose standard.

Download dataset

Size

2,004 images + JSON COCO keypoints, 4 classes, ~508 MB

Licence

CC BY 4.0

Description

‍

The Tennis Player Actions Dataset brings together more than 2,000 images classified into 4 tennis actions: forehand, backhand, serve and waiting position. Each image is annotated in COCO format with 18 key body points (according to OpenPose), making it an excellent medium for analyzing human movement in sport.

‍

What is this dataset for?

‍

Training models for the recognition of sports actions from still images
Analyze body movements using fine keypoint annotation
Develop coaching or automatic scoring assistants in tennis

‍

Can it be enriched or improved?

‍

Yes, you can add other viewpoints, integrate more specific actions, or cross-reference this data with video sequences to improve the performance of temporal models. The quality of the annotations also allows enrichment by 3D modeling or augmented generation.

‍

🔎 In summary

Criterion	Evaluation
🧩 Ease of use	⭐⭐⭐⭐⭐ (Clear structure, standard COCO annotations)
🧼 Need for cleaning	⭐⭐⭐⭐⭐ (Very low – ready-to-use)
🏷️ Annotation richness	⭐⭐⭐⭐⭐ (Excellent – 18 keypoints annotated per image)
📜 Commercial license	✅ Yes (CC BY 4.0)
👨‍💻 Beginner friendly	⚡ Yes – good entry point for computer vision projects
🔁 Fine-tuning ready	🖼️ Excellent for fine-tuning on human action recognition
🌍 Cultural diversity	⚠️ Limited – mainly Asian data (Taiwanese universities)