ECG-TransUNet v1.0.0 - Scientific Data submission release

Open-source training code for ECG-TransUNet, an ECG heartbeat annotation and correction network. The workflow consists of three model branches: base, gan, and gan-base. Each branch is trained in two stages: initial training on the public ECGModelTrainData corpus and correction fine-tuning on the curated excellent_data.hdf dataset. Final R-peak predictions are generated by joint inference across the three trained branches rather than by using a single model output. This release provides the public software snapshot used for the Scientific Data submission. Related repositories: SwineSync-OpenSource and OpenCalori-Swine. Paired ScienceDB dataset DOI for the January-2025 training cohort: 10.57760/sciencedb.35878. License: MIT.