3D dynamic synchrotron images of sandstone rock and trained weights of the P3T-Net

These are the dataset and trained models to demo the ability of P3T-Net, including:

A trained model for the semantic indication module, which uses a U-ResNet architecture: Semantic_Indication_Modules.pt

A trained model for the domain transfer module, which uses a CycleGAN-type architecture: Domain_transfer.pt

A trained model for misalignment fixing, which uses a GAN-based architecture: Misalignment_Fixing_Module.pt

A noisy synchrotron scan of a sandstone image under a core flooding condition.

These are the pretrained weight that can be directly applied to transfer the dynamic synchrotron sandstone image to a clean version. The source code of P3T-Net can be found: https://github.com/KunningTang1/P3T-Net-for-3D-large-image-transfer.git