1from huggingface_hub import hf_hub_download
2
3# Download supervised model
4model_path = hf_hub_download(
5 repo_id="tyoung089/ECG_overread",
6 filename="supervised/best.pt"
7)
8
9# Load model
10import torch
11checkpoint = torch.load(model_path, map_location="cpu")
1# Clone the code repository
2git clone https://github.com/YOUR_USERNAME/ecg-overread.git
3cd ecg-overread
4
5# Download all weights
6pip install huggingface_hub
7python download_weights.py --outdir ./weights
8
9# Run evaluation
10python Supervised/ecg_eval_cls.py \
11 --csv /path/to/test.csv \
12 --ckpt ./weights/supervised/best.pt \
13 --label-prefix "label_diag__"
1{
2 "model": state_dict, # Model weights
3 "opt": optimizer_state, # Optimizer state
4 "sched": scheduler_state, # Scheduler state
5 "epoch": int, # Training epoch
6 "best": float, # Best validation metric
7 "classes": list, # Class names
8}
1{
2 "model": state_dict, # Full model (ECG + Text encoder)
3 "epoch": int,
4 "args": training_args,
5}
1@inproceedings{overread2026midl,
2 title={Beyond Machine Interpretation: Learning from Expert Over-Reads Improves ECG Diagnosis},
3 author={Kwak et al.},
4 booktitle={Under Review for Medical Imaging with Deep Learning (MIDL)},
5 year={2026}
6}
