As you know, AI can paint:
What if we want AI to paint a real turbulant flame field? This make sense!
Install the required enviroment:
conda create -f environment.yml
The training process is based on pytorch and pytorch-lightning. All the configurations are stored in the config folder and will be arranged by hydra.
Before running the training process, modify the config/default_training.yaml file to fit your own data path. cache_path should be the directory where T1030 and similar folders are stored. For more information about the parameters in the config, look at the xxx_training.py for what parameters are used. Also, you may need to modify the variable devices in the training python script to fit your hardware. The experiments are conducted on 8 * 3090 GPUs.
To train the autoencoder model, run the following command:
python encoder_training.py exp_name=autoencoder trainer.max_epochs=100
To train the flame model, run the following command:
python vit_training.py exp_name=vit trainer.max_epochs=10000