Dataset for: Markerless Video-Based Gymnastics Score Prediction Using Biomechanical Feature Attention

This repository contains the processed datasets and analysis code associated with the study “Action-Aware Biomechanical Feature Attention for Markerless Gymnastics Score Prediction.” This new version updates the processed dataset and analysis code compared with Version 1.0. The dataset includes 500 valid video-derived gymnastics sequences, updated biomechanical feature files, revised execution-score labels assigned by certified judges, and scripts for feature extraction, preprocessing, model training, and evaluation. Raw video recordings are not publicly shared due to privacy and ethical considerations. The released data consist exclusively of processed pose representations and derived biomechanical features to support methodological transparency and reproducibility without exposing identifiable participant information.