Name		Name	Last commit message	Last commit date
parent directory ..
train_crnn_cinc2020		train_crnn_cinc2020
train_crnn_cinc2021		train_crnn_cinc2021
train_crnn_cinc2023		train_crnn_cinc2023
train_hybrid_cpsc2020		train_hybrid_cpsc2020
train_hybrid_cpsc2021		train_hybrid_cpsc2021
train_mtl_cinc2022		train_mtl_cinc2022
train_multi_cpsc2019		train_multi_cpsc2019
train_unet_ludb		train_unet_ludb
.gitignore		.gitignore
README.md		README.md

README.md

benchmarks of CinC and CPSC challenges, and using other databases

a large part are migrated from other DeepPSP repositories, some are implemented in the old fasion, being inconsistent with the new system architecture of torch_ecg, hence need updating and testing

Benchmark	Architecture	Source	Finished	Updated	Tested
CinC2020	CRNN	DeepPSP/cinc2020	✔️	✔️	✔️
CinC2021	CRNN	DeepPSP/cinc2021	✔️	✔️	✔️
CinC2022¹	Multi Task Learning (MTL)	DeepPSP/cinc2022	✔️	✔️	✔️
CinC2023¹	CRNN	DeepPSP/cinc2023	✔️	✔️	❌
CPSC2019	SequenceTagging/U-Net	NA	✔️	✔️	✔️
CPSC2020	CRNN/SequenceTagging	DeepPSP/cpsc2020	✔️	❌	❌
CPSC2021	CRNN/SequenceTagging/LSTM	DeepPSP/cpsc2021	✔️	✔️	✔️
LUDB	U-Net	NA	✔️	✔️	✔️

Known Issues

Slicing data for CPSC2021 is too slow. An offline generated (sliced) dataset is hosted at Kaggle.
Dataset for LUDB is too slow
cli for training models are completely not tested

Although CinC2022 dealt with acoustic cardiac signals (phonocardiogram, PCG), CinC2023 dealt with electroencephalogram (EEG) signals, the tasks and signals can be treated similarly. ↩ ↩²