Effective Training Strategies for Deep Learning-based Precipitation Nowcasting and Estimation

In this software, we provided the following implementations in PyTorch:

• Model and Dataset implementation (model.py and dataset.py): U-Net based model (1x1 resolution) for precipitation nowcasting and estimation, and the dataset for each target task.
• Codes for experiments: Codes for pre-training and fine-tuning the model, and evaluating the performance of the trained model.

General Information

• Authors: Jihoon Ko*, Kyuhan Lee*, Hyunjin Hwang*, Seok-Geun Oh, Seok-Woo Son, Kijung Shin
• Version: 1.0

In order to download example checkpoints for running demos, use the following commands:

$ wget https://www.dropbox.com/s/oitxstngis1ht1d/pretraining_10.pkt -O precipitation\ nowcasting/checkpoints/pretraining_10.pkt
$ wget https://www.dropbox.com/s/p3062yite6lvnq3/finetuned.pkt -O precipitation\ nowcasting/example_checkpoints/finetuned.pkt

For more details, see user_guide.pdf.