MGDM

This repository contains the official implementation of the paper:
Diffusion-Based Surrogate Modeling and Multi-Fidelity Calibration

---

🎥 Generated Videos

Fluid Dynamics

Laser-Based Metal Additive Manufacturing (LBMAM) Process

You can also generate the videos locally.

First, the zipped pretrained models should be unzipped by running

unzip models/*.zip

Then, the video frames are generated by running:

python make_heat_video.py

---

📦 Dataset

The fluid_data_gen folder contains simulation code based on fluidsim.

To download pre-generated and preprocessed datasets from Zenodo, run:

bash download_fluid_dataset.sh

---

🧠 Training

To train the diffusion-based denoising U-Net models, simply run:

python diffusion_fluid.py

The model checkpoints will be automatically saved under the /models folder.

📁 File Organization

• ./fluid_data_gen folder contains another repository that generates fluid simulations.

• ./generated_video saves two sample videos rendered on the website.

• ./models contains model weights of deep denoising neural networks.

• ./playground contains the implementation that solves 2D heat equation and estimates parameters by mse.

• ./videosamples contains video frames generated by DBS.

• diffusion_fluid.py implements DBS training and sampling on the fluid dataset.

• diffusion_heat.py implements DBS training and sampling on the LBMAM dataset.