Article title: A multi-condition diffusion model controls the reconstruction of 3D digital rocks
Journal title: Computers and Geosciences
python+pytorch
GPU: RTX 3060 or 3060+
pip install -r requirement.txt
We provide pre-trained models for heterogeneous carbonate rocks, and if you want to try to generate digital cores, you can run MainCondition_new.py file while the pre-trained model is placed in the CheckpointsCondition folder.
The generated result will be saved in the npydata folder.
Please run MainCondition_new.py and change the 'state' to 'train' and 'path' to the location of your dataset in MainCondition_new.py.
Regarding the training parameters in MainCondition_new.py, 'epoch' represents the number of training iterations, 'batch_size' refers to the number of training batches, and 'T' represents the time step in the diffusion model equation, typically set to 1000. A value of 500 will result in lower resolution effects. 'channel' represents the number of channels to adjust based on hardware requirements. 'label' corresponds to the porosity parameter, 'labelA' represents the average pore diameter, and 'labelB' represents the standard deviation of pore diameter. For a description of other parameters, please refer to the paper.
Although we have proposed a relatively new approach, the initial idea and design were inspired by (https://video-diffusion.github.io/), and the code structure was inspired by （https://github.com/zoubohao/DenoisingDiffusionProbabilityModel-ddpm-/tree/main）
meanderpy is licensed under the Apache License 2.0.