Lagrangian Neural Style Transfer for Fluids

Tensorflow implementation of Lagrangian Neural Style Transfer for Fluids.

Byungsoo Kim, Vinicius C. Azevedo, Markus Gross, Barbara Solenthaler

Computer Graphics Laboratory, ETH Zurich

(Note that Transport-Based Neural Style Transfer for Smoke Simulations (TNST) implementation is moved to tnst branch.)

Requirements

This code is tested on Windows 10 with GTX 1080 (8GB) and the following requirements:

• Python 3
• TensorFlow 1.15
• mantaflow
• SPlisHSPlash
• Partio

Run setup.bat for setup. (3rd parties must be installed manually).

Also download the pre-trained inception and vgg19 networks, and unzip it in data/model.

Usage

For details about parameters and examples, please take a closer look at run.bat and corresponding demo code, and for semantic transfer, see this page for pattern selection.

Results (smokegun, single frame)

Results (chocolate, sequence)

Results (dambreak2d, sequence)

Author

Byungsoo Kim / byungsook@github