Visual Prediction of Typhoon Clouds with Hierarchical Generative Adversarial Networks

This is a tensorflow implementation of visual prediction of typhoon clouds with hierarchical generative adversarial networks.

Requirements:

Python 2.7

Tensorflow 1.2.0

You can download the pretrained model here https://pan.baidu.com/s/1Eyanke10-h75vamsWK5MYQ. The password is vtmk. The downloaded three files should be put in 'HGAN-TyphoonCloudVisualPrediction/Save/Models/'.

Test the model:

python avg_runner -n name -l (xx/xx)/HGAN-TyphoonCloudVisualPrediction/Save/Models/model.ckpt -T

Train a new model:

python avg_runner.py -n name -s steps

Training data:/Data/Train/xx/

Testing data:/Data/Test/xx/

The models trained and generated images will be saved in '/Save/..'

Results demos

• Demo 1 ( Typhoon Morakot ) :

The following demo shows the prediction of continuous typhoon cloud movements in ten time steps of the typhoon Morakot.

The following dynamic map shows the optical flow for the predicted typhoon cloud sequence. The optical flow indicates the direction and size of typhoon cloud movements.

• Demo 2 ( Typhoon Maria ) :

The following demo shows the prediction of continuous typhoon cloud movements in ten time steps of the typhoon Maria.

The following dynamic map shows the optical flow for the predicted typhoon cloud sequence.

We acknowledge the authors for the code released at ''https://github.com/dyelax/Adversarial_Video_Generation'', which provides useful operations for our implementation.