VortexTransformer-Fitted Vatistas Velocity Profile

Description

This dataset contains the training and validation data used in the paper “VortexTransformer: End-to-End Objective Vortex Detection in 2D Unsteady Flow Using Transformers.”

Dataset Generation

The dataset was generated by fitting flow-field patches from the following datasets using simulated annealing followed by gradient-based optimization:

• cylinder2d

• Heated Cylinder with Boussinesq

• Rotation Four Center

• beads_WeinkaufTheisel2010

• pipecylinder2d(Cylinder Flow Around Corners)

• doublegyre2d

Each patch is represented using parametrized Vatistas vortex models. The dataset also includes the noise-based augmentations at different noise levels described in the paper. All the above flows' original data can be found at  https://vc.tf.fau.de/publications/datasets/

Purpose

Conventional vortex-segmentation criteria, such as the Instantaneous Vorticity Deviation (IVD), typically require users to manually specify parameters such as segmentation thresholds and local neighborhood sizes. These parameter choices may significantly affect the resulting vortex regions.

To provide analytically defined vortex labels, this dataset represents flow fields using parametrized Vatistas vortex models. The corresponding vortex-core regions can then be derived directly and analytically from the fitted Vatistas parameters. These analytically generated regions serve as training and validation labels for learning-based vortex segmentation.

Example Code

Sample Python code for loading this dataset and training the VortexTransformer model is available in train.py:

https://github.com/Cindy-xdZhang/PyflowVis/blob/main/train.py

Citation

If you use this dataset or any of its components in your research, please cite the following paper:

@article{Zhang2025VortexTransformer,
  author  = {Zhang, X. and Rautek, P. and Hadwiger, M.},
  title   = {VortexTransformer: End-to-End Objective Vortex Detection in 2D Unsteady Flow Using Transformers},
  journal = {Computer Graphics Forum},
  volume  = {44},
  number  = {2},
  pages   = {e70042},
  year    = {2025},
  doi     = {10.1111/cgf.70042},
  url     = {https://onlinelibrary.wiley.com/doi/abs/10.1111/cgf.70042}
}