Both. The paper mentions using a deblurrer and novel view synthesis model(ExBluRF).