Ed He Blog

Be what you want.

An end-to-end framework for travel time estimation - DeepTTE

When Will You Arrive? Estimating Travel Time Based on Deep Neural Networks

DeepTTE DeepTTE is comprised of three components: attribute component, spatio-temporal learning component, and multi-task learning component. The attribute component is used to processes the exter...

AlexNet笔记

让卷积神经网络重新进入大众视野

AlexNet是Hinton的学生Alex提出的一个卷积神经网络,最早用于ILSVRC-2010的比赛中,轰动一时。其中包括了五个卷积层和三个全连接FC层,迫于当时显卡的显存限制,在训练的时候使用了两块Geforce GTX 580 3GB显卡,其有如下特点: 使用Overlapping Maxpooling代替Mean Pooling,避免其带来的模糊效果; 使用ReLU作为激...

1x1 Convolution in Neural Network

How it works

最近在阅读Network in Network和GoogLeNet原文时,发现两篇文章都重点介绍了1x1卷积的使用。 其实1x1卷积操作与普通的卷积没什么区别,只是卷积核大小为1x1,它的主要作用有: 降维和升维(参考ResNet); 在不改变feature map大小的基础上增加非线性,增加模型深度以提高模型的表达能力。 On the other hand,1x1卷积就是在对...

An Introduction of Person Re-identification

draft version

做了一段时间的person re-identification,一直想写个综述,趁现在还记得,赶紧写点东西。person re-ID中文名可以叫做行人重识别,近几年在CV各大会议上都是十分火爆,而且随着deep learning在这个领域的运用越来越成熟,其识别率也越来越高,虽然在实际应用中还会有各种各样的问题,但最起码我们已经可以看到这个领域的逐渐成熟。我会根据我自己的理解对这个领域做个简...

mAP in Person Re-ID

How to calculate mAP

Introduction Mean average precision (mAP) has been widely used in object detection and image retrieval tasks, especially the person re-identification task. Here I will illustrate the calculation of...

C++ const 用法解析

浅谈各种情况下的const用途

Introduction 在C++中,const的使用十分令人confusing。一般来说,const修饰哪个部分,直接看其左边的,如果左边没东西,就看其右边的。 Basic int const a = 8; const int a = 8; 最基本的用法就是直接修饰一个变量,使其成为常变量,必须在定义时初始化,且后续不能修改其值。 int const * a = 8; const i...

Ensembling Methods in Machine Learning

Bagging and Boosting

Introduction Ensemble methods are meta-algorithms that combine several machine learning techniques into one predictive model in order to decrease variance(bagging), bias(boosting), or improve pred...

C++ Notes

updating

Author:Xiang He August 28, 2018 1 Basic ideas 1 e x p l i c i t Account ( std : : s t r i n g accountName ) 2 : name{accountName} { // member i n i t i a l i z e r 3 // empty body 4 } Construct...