Skip to content

Using the Pytorch to build an image temporal prediction model of the encoder-forecaster structure, ConvGRU kernel & ConvLSTM kernel

Notifications You must be signed in to change notification settings

chencodeX/RNN_Pytorch

Folders and files

NameName
Last commit message
Last commit date

Latest commit

author
zihao.chen
Apr 30, 2019
7da4ce7 · Apr 30, 2019

History

11 Commits
Apr 20, 2018
Apr 20, 2018
Sep 10, 2018
Apr 20, 2018
Apr 20, 2018
Apr 30, 2019
Apr 27, 2018
Apr 20, 2018
Apr 20, 2018
Apr 20, 2018

Repository files navigation

RNN_Pytorch

使用pytorch框架搭建一个encoder-forecaster结构的图像时序预测模型,ConvGRU内核

##文件说明

BMSELoss.py 一个对不同降雨级别进行加权损失的loss函数。
ConvGRUCell.py 利用卷积操作实现的ConvGRU内核,单Cell
ConvLSTM.py 利用卷积操作实现的ConvLSTM内核,包含了单Cell,多层Cell的实现方式,以及一个实验用的RNN模型
encoder.py 序列编码结构
forecaster.py 序列预测结构
HKO_EF.py 尝试一个试验性质的训练方式
HKO_model.py HKO-7模型的搭建和训练
RNN.py 一般的RNN模型
RNN_train.py RNN模型的训练

##模型说明

本项目参考的HKO模型为论文  [Deep learning for precipitation nowcasting: A benchmark and a new model](http://papers.nips.cc/paper/7145-deep-learning-for-precipitation-nowcasting-a-benchmark-and-a-new-model)
encoder和forecaster 结构皆参照此论文编写
本文的数据样本原始大为20*1*477*477,训练过程中会缩放或随机切割到120*120大小。
原始数据为0-70之间的DBz值,在输入网络前已经映射到(0-255)/255.所以代码中的BMSELoss的阈值也经过了调整。

LICENSE

About

Using the Pytorch to build an image temporal prediction model of the encoder-forecaster structure, ConvGRU kernel & ConvLSTM kernel

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages