guillherms/human-activity-pose_v3

annotations_creators:

• —human-annotated language:
• —en license: mit prettyname: Human Activity Pose Dataset (Split Version) taskcategories:
• —pose-estimation
• —action-recognition task_ids:
• —human-activity-recognition size_categories:
• —1K<n<10K --- # 🧍 Human Activity Pose Dataset

This dataset contains pose landmarks extracted using MediaPipe Pose, annotated with activity labels and English descriptions.

## 📊 Dataset Structure

• —train/ → 80% of the samples for training
• —validation/ → 20% of the samples for validation

## 🧠 Each record contains:

• —33 pose keypoints (x, y, z, visibility)
• —label: activity name (e.g., reading, dancing, office_work)
• —description: textual description of the action context.

## 🧩 Source Dataset created by Guilherme Santos using MediaPipe and OpenCV.