摘要
行人属性识别,是一个多标签分类任务。Paper With Code主页链接如下:
Pedestrian Attribute Recognition
从上图我们介意得知,常用的数据集有PA-100K、PETA、RAP、UAV Human等。
1、PA-100K数据集
PA-100K数据集是迄今为止用于行人属性识别的最大数据集,其中包含从室外监控摄像头收集的总共100000张行人图像,每张图像都有26个常用属性。根据官方设置,整个数据集随机分为80000个训练图像、10000个验证图像和10000个测试图像。
下载链接:
链接:https://pan.baidu.com/s/1Gjvg920nBrXFCiAQUrmbnA
提取码:8dq6
行人属性
总共26个属性,如下图:
在这里插入代码片- 性别:男、女
- 年龄:小于18、18-60、大于60
- 朝向:朝前、朝后、侧面
- 配饰:眼镜、帽子、无
- 正面持物:是、否
- 包:双肩包、单肩包、手提包
- 上衣风格:带条纹、带logo、带格子、拼接风格
- 下装风格:带条纹、带图案
- 短袖上衣:是、否
- 长袖上衣:是、否
- 长外套:是、否
- 长裤:是、否
- 短裤:是、否
- 短裙&裙子:是、否
- 穿靴:是、否
标签:
提取mat里面的信息
import pandas as pd
import scipy
from scipy import io
data = scipy.io.loadmat('annotation.mat')
def mat2txt(data, key):
subdata = data[key]
dfdata = pd.DataFrame(subdata)
dfdata.to_csv("%s.txt" % key, index=False)
if __name__ == "__main__":
data = scipy.io.loadmat("annotation.mat")
key_list = ["attributes", "test_images_name", "test_label",
"train_images_name", "train_label",
"val_images_name", "val_label"]
for key in key_list:
mat2txt(data, key)
2、PETA 数据集
PETA (PEdesTrian Attribute)数据集包含了8705个行人,共19000张图像(分辨率跨度范围大,从17x39到169x365的大小都有)。每个行人标注了61个二值的和4个多类别的属性。实际上,PETA数据集是由多个较小的行人重识别数据集经过属性标注后汇集而成的。如下图·:
该数据集的缺点是对同一个人的不同图像标注完全相同的属性,即便在某些区域不可见的情况下,依然保持属性不变(如在鞋子被遮挡的情况下,仍然对该图像标注了鞋子的信息),目前用到该数据集的时候,都是从中选取35个属性:
数据集下载
官网链接:http://mmlab.ie.cuhk.edu.hk/projects/PETA.html
行人属性
61个二值属性:
2 accessoryHeadphone
4 personalLess15
5 personalLess30 Age16-30
6 personalLess45 Age31-45
7 personalLess60 Age46-60
8 personalLarger60 AgeAbove60
9 carryingBabyBuggy
10 carryingBackpack Backpack
11 hairBald
12 footwearBoots
13 lowerBodyCapri
14 carryingOther CarryingOther
15 carryingShoppingTro
16 carryingUmbrella
17 lowerBodyCasual Casual lower
18 upperBodyCasual Casual upper
19 personalFemale
20 carryingFolder
21 lowerBodyFormal Formal lower
22 upperBodyFormal Formal upper
23 accessoryHairBand
24 accessoryHat Hat
25 lowerBodyHotPants
26 upperBodyJacket Jacket
27 lowerBodyJeans Jeans
28 accessoryKerchief
29 footwearLeatherShoes Leather Shoes
30 upperBodyLogo Logo
31 hairLong Long hair
32 lowerBodyLongSkirt
33 upperBodyLongSleeve
35 lowerBodyPlaid
37 lowerBodyThinStripes
38 carryingLuggageCase
39 personalMale Male
40 carryingMessengerBag MessengerBag
41 accessoryMuffler Muffler
42 accessoryNothing No accessory
43 carryingNothing No carrying
44 upperBodyNoSleeve
45 upperBodyPlaid Plaid
46 carryingPlasticBags Plastic bag
47 footwearSandals Sandals
48 footwearShoes Shoes
49 hairShort
50 lowerBodyShorts Shorts
51 upperBodyShortSleeve ShortSleeve
52 lowerBodyShortSkirt Skirt
53 footwearSneaker Sneaker
54 footwearStocking
55 upperBodyThinStripes Stripes
56 upperBodySuit
57 carryingSuitcase
58 lowerBodySuits
59 accessorySunglasses Sunglasses
60 upperBodySweater
61 upperBodyThickStripes
62 lowerBodyTrousers Trousers
63 upperBodyTshirt Tshirt
64 upperBodyOther UpperOther
65 upperBodyVNeck V-Neck
4个多类别属性:
- footwear: Black, Blue, Brown, Green, Grey, Orange, Pink, Purple, Red, White, Yellow
- hair: Black, Blue, Brown, Green, Grey, Orange, Pink, Purple, Red, White, Yellow
- lowerbody: Black, Blue, Brown, Green, Grey, Orange, Pink, Purple, Red, White, Yellow
- upperbody: Black, Blue, Brown, Green, Grey, Orange, Pink, Purple, Red, White, Yellow
转载:https://blog.csdn.net/hhhhhhhhhhwwwwwwwwww/article/details/128644737