2018雲從人頭計數

the first column	the second column
linux	keypoints
ctrl shift +	unbuntu終端放大
ctrl -	unbuntu終端縮小

1 WACV2019

• Crowd Counting Using Scale-Aware Attention Networks
  • 尺度感知注意力

2 AAAI2019

• Almost Unsupervised Learning for Dense Crowd Counting

  • 無人監督的密集人群計數學習

• ls是單詞list的簡寫，Linux有些命令的名字大多采用單詞簡寫，比如rm命令是remove的簡寫，而mv命令則是move的簡寫，還有cp=copy，chown=change owner等等。

在神經網絡中，我們知道卷積層通過卷積核和原始特征的線性結合得到輸出特征，由於卷積核通常是局部的，為了增加感受野，往往采取堆疊卷積層的方式，實際上這種處理方式並不高效。同時，計算機視覺的很多任務都是由於語義信息不足從而影響最終的性能。自注意力機制通過捕捉全局的信息來獲得更大的感受野和上下文信息。

command line arguments:

命令行參數(這定義賊TM神奇)

在歐式空間中，這個問題顯然是“只有一條”，然而在非歐空間中，這個問題就不一定了。
馬鞍形的雙曲拋物面，就是一個非歐空間的例子。現實中，非歐空間的例子還很多。
下圖中，粉色和黃色線都平行於藍線，且都經過同一點。更重要的是，藍色和黃色的平行線並非
與歐式空間中的平行線一樣處處距離相等。

Character encoding

encoding

春眠不覺曉，處處聞啼鳥。

春眠不覺曉，處處聞啼鳥。

The First Column	The Second Column
sys.argv[]	list
argv	argument vector
PIL	Python Image Library
resize	img = img.resize((width , height),Image.ANTIALIAS)
antialias	[平滑，抗鋸齒](https://www.cnblogs.com/hzm112567/p/4155167.html）
normal distribution	\(f(x;\mu,\sigma)=\frac{1}{\sqrt{2\pi}\sigma}exp{-\frac{(x-\mu)^2}{2\sigma^2}}\)
import base64	Base64 is a way of encoding invisible characters into characters
UTF-8	Unicode Transformation Format
Byte	字節
Character encoding	字符編碼
encoding	編碼

#rotating image
w = img.shape[1]
#convert picture to word
with open("C:\\Users\\user\\Desktop\\111.png

The first Column	The Second Column
FocalLoss	one-stage
Binary classification	二分類
Recall	TP/(TP+FN)
SSD	Single Shot MultiBox Detector
SSD:One-Stage	one-stage improve detection speed
Recall	FN 真故障
mAP	mean Average Precision
PR curve	Precision-Recall,AP就是該曲線下的面積
IoU threshold	交並補閾值_一般0.5
The PASCAL Visual Object Classes	VOC
NMS	[
AP	average precision
binary classification cross entorpy	CE
NMS	non-max suppression algorithm
jieba tfidf	Top20

transmission to competition

transmission

the first column	the second column
IoU	Intersection over Union
IoU	\(\div{Area of Overlap}{Area of Union}\)
Overlap	重疊部分
union	聯合
intersection	相交
ground truth bounding boxes
置信框	置信度
prior box	Be similiar to Prior Box
SSD	Single Shot MultiBox Detector
one-hot	label and one-hot
normalization	batch normalization
normalization	歸一化
tanh(hyperbolic tangent)	hyperbolic 雙曲線的
tangent	正切
mean	平均數
batch normalization instance normalization	BN and IN
Standardization	standardization
Discretization	離散化
PCA	principal component Analysis
CVR	Conversion Rate

2017_daguan_cup

the first column	the second column
1	1
2	2
3
Android development route	going to Android
Android development needs	2018 招聘
leetcode	leetcode
lintcode	lintcode

IoU threshold

(1)produce many object proposals

when the IoU threshold is small,such as 0.1,the result is that two boxes will become one box

when the IoU threshold is large,such as 0.6,so that getting the wrong result

wdir

working directory