• PyTorch version >= 1.5.0
• Python version >= 3.6
• To preprocess ECG datasets: pip install scipy==1.3.1
• For large datasets install PyArrow: pip install pyarrow

Before pre-processing, please ensure that your data directory structure is like:

path/to/ECG_digitized
├─ F00001/
├─ ...
└─ F97801
   ├─ lead_0.png
   ├─ ...
   └─ lead_12.png

Given the root directory, ECG_digitized, containing ECG images (.png) and path to ECG_Datasheet_sess.csv, run:

$ python preprocess/preprocess.py \
    /path/to/ECG_digitized \
    --dest /path/to/data \
    --csv-path /path/to/ECG_Datasheet_sess.csv

It will convert ECG images (.png) to waveform data (.mat), and output them into --dest directory.

Note that it might take around 3 hours totally.

Given the root directory containing pre-processed data, run:

$ python preprocess/manifest.py \
    /path/to/data \
    --dest /path/to/manifest \
    --valid-percent $N

$N should be set to reasonble number (<= 0.5) to split the dataset into train, valid, and test set.

$ CUDA_VISIBLE_DEVICES=$N python train.py \
    --data /path/to/manifest \

Note: set $N to denote what gpus you will use to train the model.
It will train the model with default configurations.

If you want to train model with your own configurations, add appropriate arguments to the command.
The available arguments are described in train.py