DDPrism is a JAX-based framework for solving multi-view source separation problems. It iteratively learns diffusion model priors which enable direct sampling from the joint posterior without making any explicit source assumptions. More details on the problem, the method, and the results can be found in this paper.

DDPrism requires some additional dependencies to run the experiments:

For basic usage of DDPrism's core functionality (diffusion models, sampling, and utilities):

git clone https://github.com/swagnercarena/ddprism.git
pip install .

For running experiments, training models, and using all features:

git clone https://github.com/swagnercarena/ddprism.git
pip install .[experiments]

or equivalently:

git clone https://github.com/swagnercarena/ddprism.git
pip install .[all]

For development work:

git clone https://github.com/swagnercarena/ddprism.git
pip install -e .[experiments]