Semantic Diffusion Posterior Sampling for Cardiac Ultrasound Dehazing

Tristan Stevens  |  Oisín Nolan  |  Ruud van Sloun

Eindhoven University of Technology, the Netherlands

Installation

The algorithm is implemented using Keras with JAX backend. Furthermore it heavily relies on the zea ultrasound library.

Either install the following in your Python environment, or use the Dockerfile provided in this repository.

# requires Python>=3.10 environment
pip install -r requirements.txt

Also install JAX. Note this can vary depending on your system.

pip install jax[cuda12]

Note

Although the code was primarily tested with JAX as the Keras backend, TensorFlow and PyTorch should also work.

Running the algorithm

Some example images are downloaded in the ./assets folder. The models are automatically downloaded from the Hugging Face Model Hub.

python main.py --input-folder ./assets --output-folder ./temp

Alternatively, you can use the Gradio app provided in this repository to interact with the model via a web interface. To launch the app, run:

python app.py