Skip to content
pytorch implemention of trajGRU.
Branch: master
Clone or download
Latest commit 1e38d21 Jun 22, 2019
Type Name Latest commit message Commit time
Failed to load latest commit information.
experiments Update benchmark May 2, 2019
hko_data Dataload, Benchmark, ConvLSTM and EF model Apr 11, 2019
nowcasting fix convlstm bug Jun 22, 2019
demo.gif README Apr 26, 2019 demo and benchmark Apr 25, 2019


This repo has implemented a pytorch-based encoder-forecaster model with RNNs including (TrajGRU, ConvLSTM) to do precipitation nowcasting. For more information about TrajGRU, please refer to HKO-7.

If you are interested in my implementation of ConvLSTM and TrajGRU, please see ConvLSTM and TrajGRU. It is assumed that the input shape should be . All of my implementation have been proved to be effective in HKO-7 Dataset. Hopefully it helps your research.


Firstly you should apply for HKO-7 Dataset from HKO-7, and modify somelines in to find the dataset path. Secondly and last, run python3 experiments/trajGRU_balanced_mse_mae/, and then run python3 experiments/trajGRU_frame_weighted_mse/ since I have finetuned the model on the basis of model trained in last step.


Python 3.6+, PyTorch 1.0 and Ubuntu or macOS.



The performance on HKO-7 dataset is below.

CSI HSS Balanced MSE Balanced MAE
0.5496 0.4772 0.3774 0.2863 0.1794 0.6713 0.6150 0.5226 0.4253 0.2919 5860.97 15062.46


Dropbox Pretrained Model


    title={Deep learning for precipitation nowcasting: a benchmark and a new model},
    author={Shi, Xingjian and Gao, Zhihan and Lausen, Leonard and Wang, Hao and Yeung, Dit-Yan and Wong, Wai-kin and Woo, Wang-chun},
    booktitle={Advances in Neural Information Processing Systems},
  title={Convolutional LSTM network: A machine learning approach for precipitation nowcasting},
  author={Xingjian, SHI and Chen, Zhourong and Wang, Hao and Yeung, Dit-Yan and Wong, Wai-Kin and Woo, Wang-chun},
  booktitle={Advances in neural information processing systems},
You can’t perform that action at this time.