metadata
license: mit
Model Card for DiffuSETS
Description
- Developed by: Yongfan Lai
- License: mit
Source Code
- Repository: Github
Assets List
- prerequisites_for_inference.zip:
prerequisites
├── clf_data : Directory contains data for imbalanced classification experiments
├── clip_model.pth : CLIP model checkpoints.
├── mimic_vae_lite_0_new.pt : Subsets of VAE encoding latents of MIMIC IV ECG dataset.
├── ptbxl_text_embed.csv : Pre-generated text embeddings of PTB-XL dataset
├── ptbxl_vae.pt : VAE encoding latents of PTB-XL dataset.
├── unet_all.pth : A pretrained full capacity DiffuSETS model.
└── vae_model.pth : VAE model checkpoints suit for the unet model and latent datasets above.
- mimic_vae_lite_0_new.pt: VAE encoding latents of MIMIC IV ECG dataset. Only needed while training.
Usage
Quick Start: Download and unzip the prerequisites_for_inference.zip, place the ./prerequisites folder under repo's root, then start ECG generation through
python DiffuSETS_inference.py path/to/your/config.json
or
python -m test_scripts.diversity
For users want to train from scratch: Besides above prerequisites_for_inference.zip, the mimic_vae_0_new.pt (training data encoded by vae) should also be downloaded and put into the ./prerequisites folder.
For more details about training and inference, please refer to our github repo page