<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[weiwanshu的专栏]]></title><description><![CDATA[怕什么真理无穷，进一步有一步的欢喜]]></description><link>https://blog.youkuaiyun.com/weiwanshu</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; weiwanshu]]></copyright><item><title><![CDATA[PoseGait：A model-based gait recognition method with body pose and human prior knowledge]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/109102936</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/109102936</guid><author>weiwanshu</author><pubDate>Thu, 15 Oct 2020 18:56:05 +0800</pubDate><description><![CDATA[Related work


	基于外观的识别方法：
	
常见的输入数据形式最常见的是GEI。对GEI的处理方式一般为：

1)从视频中提取人体轮廓;
	2)通过轮廓对齐和平均计算步态能量图像GEI;
	3)然后计算两个GEIs之间的相似关系
也有直接使用剪影的方法，如Gaitset。

但是基于外观的识别方法主要的挑战是 角度 / 载荷 / 衣物 / 遮挡 等，它们会改变人的轮廓。

基于模型的方法
先建立人体的模型，根据输入数据得到模型参数，而后对模型参数变化进行分类。

难点在人体模型的定位精度以及计]]></description><category></category></item><item><title><![CDATA[gaitset程序踩坑]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/106580138</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/106580138</guid><author>weiwanshu</author><pubDate>Fri, 05 Jun 2020 23:24:46 +0800</pubDate><description><![CDATA[程序来自https://github.com/AbnerHqC/GaitSet

Pretreatment.py


报错：module 'scipy.misc' has no attribute 'imsave'


问题原因：版本不对

解决方法：pip install scipy==1.2.1 -ihttps://pypi.tuna.tsinghua.edu.cn/simple


有可能需要安装 xarray



ValueError: Object arrays cannot be loa...]]></description><category></category></item><item><title><![CDATA[【论文阅读】TCN-时间卷积网络]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/89814467</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/89814467</guid><author>weiwanshu</author><pubDate>Sat, 04 May 2019 21:54:34 +0800</pubDate><description><![CDATA[参考论文：An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling

参考链接：

序列建模之时间卷积网络TCN - 杨镒铭的文章 - 知乎 https://zhuanlan.zhihu.com/p/41987775

TCN时间卷积网络 - 剪水作花飞的文章 - 知乎...]]></description><category></category></item><item><title><![CDATA[【论文阅读】DA-RNN ＆ GeoMAN ＆ DSTP-RNN]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/89575606</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/89575606</guid><author>weiwanshu</author><pubDate>Fri, 26 Apr 2019 20:18:53 +0800</pubDate><description><![CDATA[NARX 和 注意力机制相关的论文，这三个论文的模型很像，放一起比较一下]]></description><category></category></item><item><title><![CDATA[【论文阅读】An Attentive Survey of Attention Models]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/89359793</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/89359793</guid><author>weiwanshu</author><pubDate>Wed, 17 Apr 2019 20:00:56 +0800</pubDate><description><![CDATA[Attentive Model 的优势：

1、适合multiple tasks 例如 Machine Translation, Question Answering, Sentiment Analysis, Part-of-Speech tagging, Constituency Parsing and Dialogue Systems.

2、它们被广泛地用来提高神经网络的可解释性

3、有助...]]></description><category></category></item><item><title><![CDATA[【论文阅读】Routes and clustering features of PM2.5 spillover within the 3J region at multiple timescales]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/88679663</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/88679663</guid><author>weiwanshu</author><pubDate>Wed, 20 Mar 2019 09:10:38 +0800</pubDate><description><![CDATA[Routes and clustering features of PM2.5 spillover within the Jing-Jin-Ji region at multiple timescales identified using complex network-based methods

Abstract：spillover networks at multiple timescale...]]></description><category></category></item><item><title><![CDATA[【论文阅读】Long short-term memory-Fully connected neural network for PM2.5concentration prediction]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/88670502</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/88670502</guid><author>weiwanshu</author><pubDate>Tue, 19 Mar 2019 20:03:20 +0800</pubDate><description><![CDATA[Long short-term memory - Fully connected (LSTM-FC) neural network for PM2.5concentration prediction

•The LSTM-FC neural network can give an accurate prediction of urban PM2.5contamination over the n...]]></description><category></category></item><item><title><![CDATA[【python学习】HDF数据的处理以及Pandas使用]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/88407503</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/88407503</guid><author>weiwanshu</author><pubDate>Mon, 11 Mar 2019 22:01:56 +0800</pubDate><description><![CDATA[数据读取


import pandas as pd
import numpy as np
import datetime
import pickle
df1 = pd.read_hdf('D:/000data/station_data/airemiss_station_1001A.h5')
# 得到表头信息
head = df1.columns

得到的结果——



数据准备：

1、数据切片...]]></description><category></category></item><item><title><![CDATA[【实验操作】关于深度学习中的批处理数据的问题——epochs，batch_size，iterations]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/88387739</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/88387739</guid><author>weiwanshu</author><pubDate>Mon, 11 Mar 2019 09:17:56 +0800</pubDate><description><![CDATA[深度学习框架中涉及很多参数，如果一些基本的参数如果不了解，那么你去看任何一个深度学习框架是都会觉得很困难，下面介绍几个新手常问的几个参数。

batch

深度学习的优化算法，说白了就是梯度下降。每次的参数更新有两种方式。

第一种，遍历全部数据集算一次损失函数，然后算函数对各个参数的梯度，更新梯度。这种方法每更新一次参数都要把数据集里的所有样本都看一遍，计算量开销大，计算速度慢，不支持在线学习，...]]></description><category></category></item><item><title><![CDATA[【论文阅读】PM2.5的组成成分分析]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/88374847</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/88374847</guid><author>weiwanshu</author><pubDate>Sun, 10 Mar 2019 09:20:15 +0800</pubDate><description><![CDATA[Temporal-spatial characteristics and source apportionment of PM2.5as well as its associated chemical species in the Beijing-Tianjin-Hebei region of China

We present the temp-spatial PM2.5variations...]]></description><category></category></item><item><title><![CDATA[【实验操作】Anaconda以及Spyder的一些tips]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/88309142</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/88309142</guid><author>weiwanshu</author><pubDate>Sat, 09 Mar 2019 14:13:30 +0800</pubDate><description><![CDATA[工程问题

Spyder打开一个已经存在的工程project步骤

1、project-&amp;gt;new project-&amp;gt;Existing directory-&amp;gt;Location（选择已经存在的工程文件夹）-&amp;gt;Create

2、把一个已有的spyder文件夹【.spyproject】粘到工程之中

该文件夹含有四个文件



project explorer- contro...]]></description><category></category></item><item><title><![CDATA[【论文阅读】阅读计划和总结]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/88191410</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/88191410</guid><author>weiwanshu</author><pubDate>Tue, 05 Mar 2019 17:24:14 +0800</pubDate><description><![CDATA[注意力机制PM2.5预测研究

GeoMAN，提出了一种空间局部注意力、空间全局注意力、时间注意力三层的机制【注意力机制实例】GeoMAN（一）论文阅读
	对AQI进行空间插值，设计五个直接因素子网络（历史天气、天气预报、温湿度等等），预测300+城市的48小时天气情况【论文阅读】Deep Distributed Fusion Network for Air Quality Prediction
...]]></description><category></category></item><item><title><![CDATA[【论文阅读】A Dual-Stage Attention-Based Recurrent Neural Network]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/88191128</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/88191128</guid><author>weiwanshu</author><pubDate>Tue, 05 Mar 2019 17:18:46 +0800</pubDate><description><![CDATA[摘要

Nonlinear autoregressive exogenous (NARX) model 用于predicts the current value of a time series based upon its previous values as well as the current and past values of multiple driving (exogenous)...]]></description><category></category></item><item><title><![CDATA[【论文阅读】HyperST-Net:HypernetworksforSpatio-TemporalForecasting]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/88124185</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/88124185</guid><author>weiwanshu</author><pubDate>Tue, 05 Mar 2019 16:07:48 +0800</pubDate><description><![CDATA[摘要

HyperST-Net包含三个主要模块，空间模块、时间模块和推导模块deduction module

空间模块可以得到空间特征，推导模块通过空间特征得到时间模块的权重。

原理

强调考虑和时间模块和空间模块之间的关系

1 spatial characteristics of an ST object are determined by its spatial attributes,...]]></description><category></category></item><item><title><![CDATA[托福备考相关]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/87989213</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/87989213</guid><author>weiwanshu</author><pubDate>Wed, 27 Feb 2019 20:36:45 +0800</pubDate><description><![CDATA[来源于 知乎回答

阅读

答主做Reading的方法是：不看全文直接开做。比如第一题是第一段的，那我就只看第一段，然后做第一题。接下来第二题是第二段，我才开始看第二段。答主之前试过全文看完再一口气做题，发现之前看过的内容全忘了，做题时又要重新看一遍，很浪费时间。事实证明看一段做一题的方法适合我。（当然每个人都有适合自己的方法，在平时做TPO练习的时候可以试着不断变换做题方法，找到最适合自己的）
...]]></description><category></category></item><item><title><![CDATA[【论文阅读】Geographically and temporally weighted neural networks for ground-level PM2.5]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/87968195</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/87968195</guid><author>weiwanshu</author><pubDate>Wed, 27 Feb 2019 14:42:10 +0800</pubDate><description><![CDATA[Abstract：

气溶胶光学厚度AOD数据(aerosol optical depth)和地面观测站数据构成了对PM2.5的可靠量测，一般的AOD-PM2.5关系都是使用线性模型进行模拟的。使用深度学习的方法可以模拟非线性关系，但是没有考虑空间因素。本实验使用中国地区的satellite AOD products、NDVI data 、meteorological factors 和stat...]]></description><category></category></item><item><title><![CDATA[【论文阅读】Deep Inferential Spatial-Temporal Network for Forecasting Air Pollution Concentrations]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/87965911</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/87965911</guid><author>weiwanshu</author><pubDate>Wed, 27 Feb 2019 10:34:31 +0800</pubDate><description><![CDATA[摘要：

Deep Inferential Spatial-Temporal Network以解决空间和时间方面的非线性相关问题。使用inferential predictor、spatial predictor和temporal predictor进行了48小时天气预报，主要研究数据来源于北京地区35个站点和格网气象数据

目前的主要问题：

1、多因子问题 multiple factors...]]></description><category></category></item><item><title><![CDATA[【论文阅读】Deep Distributed Fusion Network for Air Quality Prediction]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/87928506</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/87928506</guid><author>weiwanshu</author><pubDate>Tue, 26 Feb 2019 17:10:28 +0800</pubDate><description><![CDATA[Deep Distributed Fusion Network for Air Quality Prediction

摘要：基于对大气污染的领域知识，提出了一种基于深度神经网络（DNN）的方法（Deep-air），该方法由空间变换组件和深度分布式融合网络组成。考虑到大气污染物的空间相关性，组件将空间稀疏的空气质量数据转换为一致的输入，以模拟污染物源。后一种网络采用神经分布式结构，融合城市异质数据...]]></description><category></category></item><item><title><![CDATA[【注意力机制实例】GeoMAN（二）代码研究]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/87902840</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/87902840</guid><author>weiwanshu</author><pubDate>Mon, 25 Feb 2019 20:33:48 +0800</pubDate><description><![CDATA[
来源：


&quot;&quot;&quot;
Yuxuan Liang, Songyu Ke, Junbo Zhang, Xiuwen Yi, Yu Zheng
GeoMAN: Multi-level Attention Networks for Geo-sensory Time Series Prediction
27th International Joint Conference on Artificial Int...]]></description><category></category></item><item><title><![CDATA[【注意力机制实例】GeoMAN（一）论文阅读]]></title><link>https://blog.youkuaiyun.com/weiwanshu/article/details/87798893</link><guid>https://blog.youkuaiyun.com/weiwanshu/article/details/87798893</guid><author>weiwanshu</author><pubDate>Thu, 21 Feb 2019 17:27:51 +0800</pubDate><description><![CDATA[GeoMAN: Multi-level Attention Networks for Geo-sensory Time Series Prediction

摘要：使用一个考虑多传感器读数、气象数据和空间数据的，基于多层次注意的递归神经网络，预测未来几个小时内的地理传感器读数。模型包括两个方面，1) 一种动态时空依赖模型的多级注意力机制 2）一种融合不同领域外部因素的通用融合模块。主要应用在空气质...]]></description><category></category></item></channel></rss>