ecg-biometrics-bench-table-weights

ECG Biometrics Benchmark - Table-Reproduction Checkpoints

This repository contains pretrained DeepECG checkpoints required to reproduce the primary numerical benchmark tables of the ECG Biometrics Benchmark.

The release contains 390 unique checkpoints referenced by the paper-reproduction campaign. Some of these checkpoints are also reused by other analyses, but they are included here because they are required for the primary benchmark tables.

Checkpoints needed only for figures, architecture comparisons, controlled ablations, augmentation studies, or other supplementary analyses are not included unless that same checkpoint is also required by a primary table.

Contents

• 390 DeepECG checkpoints
• seven benchmark datasets
• all relevant training seeds used for table reproduction
• one sanitized metadata JSON per checkpoint
• MANIFEST.csv with model, training, protocol, integrity, and provenance information

Benchmark code

https://github.com/MParvan/ecg-biometrics-bench

Checkpoint format

Each checkpoint is a plain PyTorch state_dict.

Raw experiment sidecar files are not distributed.

The public metadata is generated from an explicit whitelist and contains no local dataset paths or user-directory paths.

Recorded source revisions

The metadata retains the Git revision recorded at training time.

Two recorded revisions in this checkpoint set, 452e88f1 and adf06bf5, were verified to contain identical source trees even though they belong to different Git history lines.

Datasets

Dataset files are not included. Users must obtain each dataset from its official source and comply with the corresponding access and licensing terms.

License

The benchmark software is distributed under the MIT License. Dataset terms remain governed by their respective providers.