implement self-attention convolution LSTM

Requirements

    torch==1.8.1
    torchvision==0.9.1
    numpy==1.19.5
    matplotlib==3.4.1

Getting Started

    git clone https://github.com/CaiCandong/sa_convLSTM.git
    cd sa_convLSTM
    python train.py

Results

模型	训练集loss	验证集loss
SA-ConvLSTM	0.0136	0.0189

Model Architecture

ConvLSTM

SA-ConvLSTM Momory Module

SA-ConvLSTM Architecture

Multi-Layer SA-ConvLSTM Architecture

References:

• https://github.com/MinNamgung/sa_convlstm
• https://github.com/hyona-yu/SA-convlstm
• Lin, Zhihui,. 《Self-Attention ConvLSTM for Spatiotemporal Prediction》
• https://github.com/jerrywn121/TianChi_AIEarth/tree/main/SAConvLSTM