FCNR: Fast Compressive Neural Representation of Visualization Images

Paper

Yunfei Lu, Pengfei Gu, Chaoli Wang

This is the official pytorch implementation for the paper "FCNR: Fast Compressive Neural Representation of Visualization Images".

Get Started

Set up a conda environment with all dependencies with Python 3.9:

pip install -r requirements.txt

Data

You can generate customized visualization images with different viewpoints and timesteps on your own dataset via volume or isosurface rendering. Here is a link to download the vortex dataset (direct volume rendering images included) we use: vortex.

Training & Inference

Specify <gpu_idx>, <exp_name> and <config_name> to start training and inferencing:

python train.py <gpu_idx> <exp_name> --config ./configs/<config_name>

An example of the configuration file we use is ./configs/cfg.json. You can follow it to implement on your customized dataset.

Results

Here is a comparison between the results of FCNR and existing baselines:

Citation

@article{lu2024fcnr,
  title={{FCNR}: Fast Compressive Neural Representation of Visualization Images},
  author={Lu, Yunfei, Gu, Pengfei, and Wang, Chaoli},
  booktitle={Proceedings of IEEE VIS Conference (Short Papers)},
  year={2024},
  note={Accepted}
}