This project applies neural style transfer techniques to ECG images for various transformations and augmentations. The repository contains scripts, models, and results for experiments conducted on ECG datasets.

• src/: Source code for the project, including transformation scripts and main experiment scripts.
• run_*.sh: Shell scripts to execute various experiments and transformations.
• data.txt: Link to download data
• requirements.txt: Python dependencies required for the project.

• run_main_merged_augmentation.sh: Train, evaluate model on both original and nst-transformed data.
• run_main_nst_augmentation.sh: Train, evaluate model on only nst-augmented data.
• run_test_all.sh: Comparative evaluation of models when only trained on either original or nst-augmented data.
• run_transform_ab_hb.sh: Apply NST to abnormal heartbeat images.
• run_transform_h_MI.sh: Apply NST to history of myocardial infarction images.
• run_transform_MI.sh: Apply NST to myocardial infarction images.
• run_transform_normal.sh: Apply NST to normal heartbeat images.

• Python 3.8 or higher

• Install dependencies using:

  pip install -r requirements.txt

1. Without Augmentation:

   sbatch run_main_no_augmentation.sh

2. With Neural Style Transfer Augmentation:

   sbatch run_main_nst_augmentation.sh

3. With Merged Augmentation:

   sbatch run_main_merged_augmentation.sh

4. Testing All Models:

   sbatch run_test_all.sh

• Special thanks to the CMSC 636 course for guidance and support.
• Neural Style Transfer techniques inspired by Gatys et al.'s work on artistic style transfer [doi: https://arxiv.org/abs/1508.06576].