[笔记] Pedestrian Attribute Recognition Dataset Summary

目录

1. PETA Dataset
2. RAP Dataset
3. PA-100K Dataset
4. WIDER Attribute Dataset
5. Database of Human Attributes (HAT)
6. Market-1501_Attribute
7. DukeMTMC-Attribute
8. Clothing Attributes Dataset
9. Parse27k Dataset
10. RAP 2.0 Dataset
11. CRP Dataset
12. APis dataset
13. Berkeley-Attributes of People dataset
14. Deepfashion dataset
15. Video-Based PAR dataset

1. PETA Dataset

链接： http://mmlab.ie.cuhk.edu.hk/projects/PETA.html

PETA Dataset由多个数据集组合而成，共有65个attribute，其中，包括61个 Binary Attribute，4个 Multi-class attributes。

2. RAP Dataset

链接： http://rap.idealtest.org/

3. PA-100K Dataset

链接： https://drive.google.com/drive/folders/0B5_Ra3JsEOyOUlhKM0VPZ1ZWR2M

4. WIDER Attribute Dataset

链接： http://mmlab.ie.cuhk.edu.hk/projects/WIDERAttribute.html

WIDER Attribute is a large-scale human attribute dataset. It contains 13789 images belonging to 30 scene categories, and 57524 human bounding boxes each annotated with 14 binary attributes.

The Binary Attributes are as follows :
Male, Long Hair, Sunglasses, Hat, Tshirt, longSleeve, Formal, Shorts, Jeans, Long Pants, Skirt, Face Mask, Logo, Plaid or Stripe.

5. Database of Human Attributes (HAT)

链接： https://jurie.users.greyc.fr/datasets/hat.html

train set size: $3500\ast 27$, test set size: $2344\ast 27$, evaluation set size: $3500\ast 27$

attribute	label
female	+1 : attribute present, -1 : attribute absent, 0 : attribute not visible or ambiguous
frontalpose	DITTO
profilepose	DITTO
turnedback	DITTO
upperbody	DITTO
standing	DITTO
runwalk	DITTO
crouching	DITTO
sitting	DITTO
armsbent	DITTO
elderly	DITTO
middleaged	DITTO
young	DITTO
teen	DITTO
kid	DITTO
baby	DITTO
tanktop	DITTO
tshirt	DITTO
mensuit	DITTO
longskirt	DITTO
shortskirt	DITTO
smallshorts	DITTO
lowcuttop	DITTO
swimsuit	DITTO
weddingdress	DITTO
bermudashorts	DITTO

6. Market-1501_Attribute

链接： https://github.com/vana77/Market-1501_Attribute

train set size: $751\ast 27$, test set size: $750\ast 27$, evaluation set size: $13115\ast 27$

attribute	representation in file	label
gender	gender	1 : male, 2 : female
hair length	hair	1 : short hair, 2 : long hair
sleeve length	up	1 : long sleeve, 2 : short sleeve
length of lower-body clothing	down	1 : long lower body clothing, 2 : short
type of lower-body clothing	clothes	1 : dress, 2 : pants
wearing hat	hat	1 : no, 2 : yes
carrying backpack	backpack	1 : no, 2 : yes
carrying bag	bag	1 : no, 2 : yes
carrying handbag	handbag	1 : no, 2 : yes
age	age	1 : young, 2 : teenager, 3 : adult, 4 : old
8 color of upper-body clothing	upblack, upwhite, upred, uppurple, upyellow, upgray, upblue, upgreen	1 : no, 2 : yes
9 color of lower-body clothing	downblack, downwhite, downpink, downpurple, downyellow, downgray, downblue, downgreen,downbrown	1 : no, 2 : yes

Smaple:

7. DukeMTMC-Attribute

链接： https://github.com/vana77/DukeMTMC-attribute

train set size: $702\ast 23$, test set size: $1110\ast 23$, evaluation set size: $13115\ast 23$

attribute	representation in file	label
gender	gender	1 : male, 2 : female
length of upper-body clothing	top	1 : short upper body clothing, 2 : long
wearing boots	boots	1 : no, 2 : yes
wearing hat	hat	1 : no, 2 : yes
carrying backpack	backpack	1 : no, 2 : yes
carrying bag	bag	1 : no, 2 : yes
carrying handbag	handbag	1 : no, 2 : yes
color of shoes	shoes	1 : dark, 2 : light
8 color of upper-body clothing	upblack, upwhite, upred, uppurple, upgray, upblue, upgreen, upbrown	1 : no, 2 : yes
7 color of lower-body clothing	downblack, downwhite, downred, downgray, downblue, downgreen, downbrown	1 : no, 2 : yes

Smaple:

8. Clothing Attributes Dataset

链接： https://purl.stanford.edu/tb980qz1002

The dataset contains 1856 images, with 26 ground truth clothing attributes collected using Amazon Mechanical Turk. All labels are arranged in the order from image 1 to 1856. Some label entries are ‘NaN’, meaning the 6 human workers cannot reach an agreement on this clothing attribute.

Details of the clothing attributes labels are shown below:

attribute	label
Necktie	1 : No necktie, 2 : Has necktie
Collar	1 : No collar, 2 : Has collar
Gender	1 : Male, 2. Female
Placket	1 : No placket, 2 : Has placket
Skin exposure	1 : Low exposure, 2 : High exposure
Wear scarf	1 : No scarf, 2 : Has scarf
Solid pattern	1 : No, 2 : Yes
Floral pattern	1 : No, 2 : Yes
Spotted pattern	1 : No, 2 : Yes
Graphics pattern	1 : No, 2 : Yes
Plaid pattern	1 : No, 2 : Yes
Striped pattern	1 : No, 2 : Yes
Red color	1 : No, 2 : Yes
Yellow color	1 : No, 2 : Yes
Green color	1 : No, 2 : Yes
Cyan color	1 : No, 2 : Yes
Blue color	1 : No, 2 : Yes
Purple color	1 : No, 2 : Yes
Brown color	1 : No, 2 : Yes
White color	1 : No, 2 : Yes
Gray color	1 : No, 2 : Yes
Black color	1 : No, 2 : Yes
Many (>2) colors	1 : No, 2 : Yes
Sleeve length	1 : No sleeves, 2 : Short sleeves, 3 : Long sleeves
Neckline	1 : V-shape, 2 : Round, 3 : Other shapes
Category	1 : Shirt, 2 : Sweater, 3 : T-shirt, 4 : Outerwear, 5 : Suit, 6 : Tank Top, 7 : Dress

9. Parse27k Dataset

链接： https://www.vision.rwth-aachen.de/page/parse27k

The attributes are all defined based on some binary or multinomial proposition. The annotated attributes include two orientation labels with 4 and 8 discretizations, and several binary attributes with an additional N/A state:

• N/A or ? - the observer cannot decide
• yes - the proposition holds
• no - the proposition does not hold

The examples are all annotated with the following attributes:

• Gender (male, female, ?)
• Posture (standing, walking, (sitting), ?)
• Orientation (4 discretizations + ?)
• Orientation8 (8 discretizations + ?)
• Bag on Left Shoulder (yes, no, ?)
• Bag on Right Shoulder (yes, no, ?)
• Bag in Left Hand (yes, no, ?)
• Bag in Right Hand (yes, no, ?)
• Backpack (yes, no, ?)
• isPushing (yes, no, ?) – child-strollers, etc.

10. RAP 2.0 Dataset

链接： https://drive.google.com/file/d/1hoPIB5NJKf3YGMvLFZnIYG5JDcZTxHph/view

11. CRP Dataset

链接： http://www.vision.caltech.edu/~dhall/projects/CRP/

12. APis dataset

链接： http://www.cbsr.ia.ac.cn/english/APiS-1.0-Database.html

13. Berkeley-Attributes of People dataset

链接： https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/shape/poselets/

train set size: $4013\ast 9$, test set size: $4022\ast 9$

attribute	label
is_male	1 : attribute present, -1 : attribute not absent, 0 : attribute unspecified.
has_long_hair	DITTO
has_glasses	DITTO
has_hat	DITTO
has_t-shirt	DITTO
has_long_sleeves	DITTO
has_shorts	DITTO
has_jeans	DITTO
has_long_pants	DITTO

14. Deepfashion dataset

链接： http://mmlab.ie.cuhk.edu.hk/projects/DeepFashion.html

• Category and Attribute Prediction Benchmark

  num of images: 289222, num of attributes:1000, num of attribute type: 50
  Note:

  1. In attribute type, “1” represents texture-related attributes, “2” represents fabric-related attributes, “3” represents shape-related attributes, “4” represents part-related attributes, “5” represents style-related attributes.
  2. In category type, “1” represents upper-body clothes, “2” represents lower-body clothes, “3” represents full-body clothes.
  3. For the clothing categories, “Cape”, “Nightdress”, “Shirtdress” and “Sundress” have been merged into “Dress”.
  4. In clothes type, “1” represents upper-body clothes, “2” represents lower-body clothes, “3” represents full-body clothes. Upper-body clothes possess six fahsion landmarks, lower-body clothes possess four fashion landmarks, full-body clothes possess eight fashion landmarks.
  5. In variation type, “1” represents normal pose, “2” represents medium pose, “3” represents large pose, “4” represents medium zoom-in, “5” represents large zoom-in.
  6. In landmark visibility state, “0” represents visible, “1” represents invisible/occluded, “2” represents truncated/cut-off.
  7. For upper-body clothes, landmark annotations are listed in the order of [“left collar”, “right collar”, “left sleeve”, “right sleeve”, “left hem”, “right hem”]; For lower-body clothes, landmark annotations are listed in the order of [“left waistline”, “right waistline”, “left hem”, “right hem”]; For upper-body clothes, landmark annotations are listed in the order of [“left collar”, “right collar”, “left sleeve”, “right sleeve”, “left waistline”, “right waistline”, “left hem”, “right hem”].

• In-shop Clothes Retrieval Benchmark

  num of images: 52712, num of items: 7982, num of attributes:463
  Note:

  1. In clothes type, “1” represents upper-body clothes, “2” represents lower-body clothes, “3” represents full-body clothes. Upper-body clothes possess six fahsion landmarks, lower-body clothes possess four fashion landmarks, full-body clothes possess eight fashion landmarks.
  2. In attribute labels, “1” represents positive while “-1” represents negative, “0” represents unknown.In attribute labels, “1” represents positive while “-1” represents negative, “0” represents unknown.
  3. In landmark visibility state, “0” represents visible, “1” represents invisible/occluded, “2” represents truncated/cut-off.
  4. For upper-body clothes, landmark annotations are listed in the order of [“left collar”, “right collar”, “left sleeve”, “right sleeve”, “left hem”, “right hem”]; For lower-body clothes, landmark annotations are listed in the order of [“left waistline”, “right waistline”, “left hem”, “right hem”]; For upper-body clothes, landmark annotations are listed in the order of [“left collar”, “right collar”, “left sleeve”, “right sleeve”, “left waistline”, “right waistline”, “left hem”, “right hem”].
  5. In clothes type, “1” represents upper-body clothes, “2” represents lower-body clothes, “3” represents full-body clothes. Upper-body clothes possess six fahsion landmarks, lower-body clothes possess four fashion landmarks, full-body clothes possess eight fashion landmarks.
  6. In variation type, “1” represents normal pose, “2” represents medium pose, “3” represents large pose, “4” represents medium zoom-in, “5” represents large zoom-in.

• Consumer-to-shop Clothes Retrieval Benchmark

  num of images: 239557, num of items: 33881, num of attributes:303, num of attribute type: 18
  Note:

  1. In attribute labels, “1” represents positive while “-1” represents negative, “0” represents unknown.
  2. In clothes type, “1” represents upper-body clothes, “2” represents lower-body clothes, “3” represents full-body clothes.
  3. In source type, “1” represents shop image, “2” represents consumer image.
  4. In variation type, “1” represents normal pose, “2” represents medium pose, “3” represents large pose, “4” represents medium zoom-in, “5” represents large zoom-in.
  5. In landmark visibility state, “0” represents visible, “1” represents invisible/occluded, “2” represents truncated/cut-off.
  6. For upper-body clothes, landmark annotations are listed in the order of [“left collar”, “right collar”, “left sleeve”, “right sleeve”, “left hem”, “right hem”]; For lower-body clothes, landmark annotations are listed in the order of [“left waistline”, “right waistline”, “left hem”, “right hem”]; For upper-body clothes, landmark annotations are listed in the order of [“left collar”, “right collar”, “left sleeve”, “right sleeve”, “left waistline”, “right waistline”, “left hem”, “right hem”].

• Fashion Landmark Detection Benchmark

  num of images: 123016, num of items: 33881, num of attributes:303, num of attribute type: 18
  Note:

  1. In clothes type, “1” represents upper-body clothes, “2” represents lower-body clothes, “3” represents full-body clothes;
  2. In variation type, “1” represents normal pose, “2” represents medium pose, “3” represents large pose, “4” represents medium zoom-in, “5” represents large zoom-in;
  3. In landmark visibility state, “0” represents visible, “1” represents invisible.

• Fashion Synthesis Benchmark

  num of images: 78979, num of items: 33881, num of attributes:303, num of attribute type: 18

15. Video-Based PAR dataset

链接： https://github.com/yuange250/MARS-Attribute

---

如果你看到了这篇文章的最后，并且觉得有帮助的话，麻烦你花几秒钟时间点个赞，或者受累在评论中指出我的错误。谢谢！

作者信息：
知乎：没头脑
LeetCode：Tao Pu
CSDN：Code_Mart
Github：Bojack-want-drink