GMT: A deep learning approach to generalized multivariate translation for scientific data analysis and visualization

PyTorch implementation of the paper GMT: A deep learning approach to generalized multivariate translation for scientific data analysis and visualization

Prerequisite

Linux
Python >= 3.7
CUDA >= 11.3
PyTorch >= 1.11.0
NumPy
RAM >= 32GB
VRAM >= 16GB

Data Format

The volume contains little-endian floats in column-major order (z-axis, y-axis, x-axis).

Training

Pretrain GMT for 500 epochs

python3 pretrain.py --data_path /your/data/path --model_path /your/model/path --max_epoch 500 --dataset dataset_name

Fine-tune GMT for 4000 epochs

python3 train.py --data_path /your/data/path --model_path /your/model/path --max_epoch 4000 --dataset dataset_name

Inference using a trained GMT model (translating 0th variable to 1st variable)

python3 inference.py --data_path /your/data/path --model_path /your/model/path --epoch 4000 --dataset dataset_name --source 0 --target 1

Citation

@article{Yao-GMT-CG23,
  title = {{GMT}: A deep learning approach to generalized multivariate translation for scientific data analysis and visualization},
  journal = {Computers \& Graphics},
  volume = {112},
  pages = {92-104},
  year = {2023},
  author = {Siyuan Yao and Jun Han and Chaoli Wang}
}

Acknowledgements