Yoga Pose Evaluation Dataset

Yoga is widely accepted globally as an effective method for regulating mind and body. However, due to the complexity of yoga poses, many beginners find it difficult to master them correctly. This leads to a high risk of injuries and improper postures without professional guidance. Most existing yoga teaching applications rely heavily on video tutorials and lack real-time feedback mechanisms, affecting the user learning experience. The Yoga Pose Evaluation Dataset aims to support the development of precise pose recognition algorithms by providing high-quality image data to correct users' posture errors in real-time. This dataset collects images using professional photography equipment under controlled lighting conditions both in studios and real yoga practice environments. Data quality is ensured through multiple rounds of annotation, real-time consistency checks, and expert reviews. The annotation team consists of over 20 members, including yoga instructors and image processing experts. Data preprocessing includes background removal, image size normalization, and color correction, stored in JPEG format, organized with a folder structure classified by pose. This dataset has several unique advantages. Annotation accuracy reaches 98%, ensuring consistency and integrity. The latest image enhancement technology and unique multi-angle collection methods are introduced to guarantee data diversity. It is widely applicable, from mobile applications to smart fitness devices, enhancing pose recognition accuracy by up to 30%. Compared to existing similar datasets, it provides more pose types and dynamic variations, and its data characteristics are global and diverse, making it more representative and valuable for research. Users can easily update the dataset to adapt to changing user needs, demonstrating strong scalability and versatility.