This is a Pytorch implementation of RAP-Net, a novel attention model for radar echo prediction (precipitation nowcasting) as described in the following paper:
RAP-Net: Region Attention Predictive Network for Precipitation Nowcasting, by Chuyao Luo, Zheng Zhang, Rui Ye, Xutao Li, Yunming Ye.
Required python libraries: torch (>=1.7.0) + opencv + numpy. Tested in ubuntu + nvidia 3090 Ti with cuda (>=11.0).
We conduct experiments on CIKM AnalytiCup 2017 datasets: CIKM_AnalytiCup_Address or CIKM_Rardar
Use any '.py' script in folds of experiment and ablation_study to train these models. To train the proposed model on the radar, we can simply run the CIKM_rap_net.py in the fold of experiment
You might want to change the parameter and setting, you can change the details of variable ‘args’ in each files for each model
The preprocess method and data root path can be modified in the data/data_iterator.py file.
These trianed models are available at: trained model
5 frames are predicted given the last 10 frames.
Besides, the trained models including RAP-Net, PredRNN and all models in the ablation study can be found in here