NeuroBiomech Experiment Data Based on Public CSL500 \/ WLASL \/ Kinetics Datasets

This dataset provides preprocessed 2D\/3D skeleton sequences derived from three public datasets \u2014 CSL500, WLASL, and Kinetics \u2014 and was used for experimental validation in the research paper \u201cNeuroBiomech: Biomechanically-Faithful Sign Language Generation via Attention-Augmented LSTM and Neural State Estimation\u201d.Each sequence has been extracted using OpenPose and refined with Kalman-filter-based smoothing and biomechanical constraint enforcement (e.g., joint angle limits, velocity continuity). The dataset is organized by source, and includes paired text annotations, filtered joint trajectories, and motion-normalized samples suitable for training and evaluating sign language generation models.This resource is intended for reproducibility and benchmarking in the fields of sign language processing, motion synthesis, and physics-constrained neural generation. Users are encouraged to cite both the original data sources and our associated publication when using this dataset.