|
|
--- |
|
|
license: apache-2.0 |
|
|
tags: |
|
|
- vit |
|
|
- ecg |
|
|
- self-supervised |
|
|
- contrastive-learning |
|
|
--- |
|
|
|
|
|
# ViT-MAE ECG Encoder |
|
|
|
|
|
This repository contains a Vision Transformer (ViT-MAE) model pretrained on ECG signals using masked autoencoding. It is a forked and modified version of the Hugging Face `transformers` library. |
|
|
|
|
|
## Model Files |
|
|
|
|
|
- `model.safetensors`: Model weights (~77MB) saved in `safetensors` format. |
|
|
- `config.json`: Model architecture and configuration. |
|
|
|
|
|
## Usage |
|
|
|
|
|
To use this model, make sure to install the forked version of `transformers` (see below), which includes modifications for original implementations of ViTMAE model from HuggingFace. |
|
|
|
|
|
```bash |
|
|
git clone git@github.com:Alsalivan/ecgcmr.git |
|
|
cd ecgcmr/external/transformers |
|
|
pip install -e . |
|
|
``` |
|
|
|
|
|
## Load model |
|
|
|
|
|
``` |
|
|
from transformers import ViTMAEModel |
|
|
model = ViTMAEModel.from_pretrained("alsalivan/vitmae_ecg") |
|
|
``` |
