Myo–Amputee & NinaPro DB3 for sEMG Hand Gesture Recognition: Dataset and Code for CNN–ViT (CViT) Comparison

This repository provides the Myo–Amputee dataset (raw and processed sEMG signals; 8 channels; 6 hand gestures; amputee participants) and reproducible code for preprocessing, training, and evaluating CNN and CNN–ViT (CViT) models for hand-gesture recognition. It includes segmentation scripts (window/overlap), subject-wise and repetition-wise train/validation/test splits, training pipelines (hyperparameters, seeds, early stopping), and metrics with 95% CIs. We also supply splits and results for comparative experiments with the public NinaPro DB3 dataset, together with statistical tests (Wilcoxon, Friedman, post-hoc where applicable) and effect-size tables. The materials enable full replication of the reported findings and extension to low- and high-channel configurations.