tfST

tensorflow implementation of the scattering transform

This implementation example runs on Hyperspectral Data.

Please cite:

@article{tfST,
	title={Three-Dimensional Fourier Scattering Transform and Classification of Hyperspectral Images},
	author={Ilya Kavalerov and Weilin Li and Wojciech Czaja and Rama Chellappa},
	journal={arXiv preprint arXiv:TBA},
	year={2019}
}

Usage

Downloading Data

See the GIC website and download for example the "corrected Indian Pines" and "Indian Pines groundtruth" datasets.

Running Classification

CUDA_VISIBLE_DEVICES=0 python main.py --dataset=IP --data_root=/scratch0/ilya/locDoc/data/hyperspec/datasets/ --train_test_splits=Indian_pines_gt_traintest.mat

Will run classification on a training set size of 10% with OA 98.30%.

Create Custom Training/Testing Splits

One training/testing split is included. Create more by editing the variables OUT_PATH, DATASET_PATH, ntrials, and datasetsamples in create_training_splits.m, and running:

matlab -nodesktop -nosplash -r "create_training_splits"

Versioning

Tested on Python 2.7.14 (Anaconda), tensorflow 1.10.1, cuda 9.0.176, cudnn-8.0. Red Hat Enterprise Linux Workstation release 7.6 (Maipo). GeForce GTX TITAN X.

License

MIT

ilyakava/tfST