vastyellownew / contournet

ContourNet

This repository contains the PyTorch implementation for ContourNet, a Convolutional Neural Network (CNN) to detect salient isocontours in XGC plasma fusion.

Dependencies

There is a conda environment (contournet.yml), however the basic requirements a pytorch, numpy, and scikit learn, as well as tqdm for progress bars.

Tested with

• pytorch 1.1.0
• scikit-learn 0.21.2
• tqdm 4.32.1
• numpy 1.16.4

Running

To run the training, simply call python Training.py. For Inference, you need to specify the model name python Inference.py <model> that has been saved by Training. The label propagation can either be run in a non-sequential manner (worse) or a sequential manner. The program parameters are the labeled time step, the slice and then a list of time step that should be propagated from the host time step. For the sequential version, several calls have to be made, e.g. python LabelPropagation.py 60 10 61, python LabelPropagation.py 61 10 62, and so on. In this case, the propagation would first go from 60 to 61, and then from 61 to 62. The non-sequential call in the following example python LabelPropagation.py 60 10 61 62 would first propagate from 60 to 61 and then from 60 to 62, which leads to a worse output quality.

Citation

Martin Imre, Jun Han, Julien Dominski, Michael Churchill, Ralph Kube, Choong-Seock Chang, Tom Peterka, Hanqi Guo, and Chaoli Wang. ContourNet: Salient Local Contour Identification for Blob Detection in Plasma Fusion Simulation Data. In Proceedings of International Symposium on Visual Computing, Lake Tahoe, NV, Oct 2019.