The source code of paper: "Experimental Study on Generative Adversarial Network for Precipitation Nowcasting"
This source code involves nine models: ConvGRU, ConvLSTM Multi Scale CNN ConvGRU GAN, ConvLSTM GAN, ConvGRU WGAN, ConvLSTM WGAN, Multi Scale GAN, Multi Scale WGAN,
The first to seventh can be run in the path of experiment/radar/ the last two can be run in the path of multi_scale_gan/avg_runner or wavg_runner/
We give the several original test samples(73 images in total) which mention in the paper. It located in the path of data/classic_data/
These data need to be preprocessed. The specific steps can be refer to read_files() function in each model file(eg. experiment/radar/convGru.py). The range of data is 0~255 and the value of 255 presents the default value. Hence, we need to filter it and all pixels whose value larger than 80.
It is difficult to train the model. Therefore, we offer the all models and you can download it following this address model_libs_link
We will keep maintaining this code recently.