meshgraphdata

This snippet converts the .tfrecord files from the meshgraphnet datasets to .vtu trajectories.

Usage

First, download a dataset using the download.sh file (copied from the meshgraphnet repository) following these steps:

./download.sh <dataset_name> <local_folder_name>

Then, convert using the data.py file:

python3 data.py <dataset_folder> <split> <output_folder>

This should create the output_folder, and store the trajectories as .vtu files in separate folders within it. In each trajectory folder, it will write a trajectory.pvd file that can be open directly with paraview to visualize the trajectory. This is a quite long process.

Examples

Issues

The metadata files for the flag_dynamic_sizing and sphere_dynamic_sizing datasets seem to have erroneous attributes for the sizing_field attributes. It is necessary to correct them by hand before converting them (issue opened here). To fix it, change this block:

"sizing_field": {
  "type": "dynamic_varlen",
  "shape": [
    -1,
     4
  ],
  "dtype": "<dtype: 'float32'>"
}

to this:

"sizing_field": {
  "type": "dynamic_varlen",
  "shape": [
    -1,
     3
  ],
  "dtype": "float32"
}