数据库
属性
- CelebA: 10,177 number of identities,202,599 number of face images, and 5 landmark locations, 40 binary attributes annotations per image.
- The ORL Database of Faces: 来自AT&T,图片有40个属性,受限场景,均匀背景幕布下采集,1992-1994
年龄、性别、人种
- MORPH: 需要$499,13673个人,共55608张。属性主打年龄,人种,性别。非受限条件采集图片。额外的还包括头发颜色(人种特征)眼镜,详细简介看这里
- UTKFace : 2W张人脸,主打人种、性别、年龄。
- CARC: 2k个名人的16W张图片,主要是年龄
- FG-NET:(82个人,共1002张),论文为2014年ECCV论文
- Adience : 2284个人,共26580张,主要为年龄、性别识别
- CACD2000 :(2000个人,共163446张)
- LAP
- IMDB-Wiki dataset: 10W个艺人,由IMDB的2W名人的46W张人脸和Wikipedia的6W张人脸组成。主要是年龄、姓名、性别(20284个人,共523051张)
- MegaAge-Asian dataset: 4W张照片,主打年龄,亚洲人脸图片。主要用来后处理的方法。
表情
- JAFFE
- CK+
- MMI
- FER-2013 Faces Database: 约3w张图片,7种表情,微软表情比赛数据
- MSFDE: 多种族7种表情
- EmotiW 一堆数据集
关键点
- BioID(1000张 20个关键点)
- LFPW:(1132 张,29个关键点)
- AFLW(25993张,21个关键点)
- COFW:(1852张,每个人脸标定29个关键点)
- ICCV13/MVFW :(2500张,68个关键点)
- OCFW: (3837张,68个关键点)
- 300-W :(600张,68个关键点)
- WFLW: 10000 faces (7500 for training and 2500 for testing) with 98 fully manual annotated landmarks
颜值
- **SCUT-FBP5500-Database: 性别和颜值,亚洲和高加索人种,5500张图片
开源项目(基本上都是多任务学习)
年龄
-
- 数据: IMDB-WIKI dataset
- 论文:[1] R. Rothe, R. Timofte, and L. V. Gool, “DEX: Deep EXpectation of apparent age from a single image,” in Proc. of ICCV, 2015. [2] R. Rothe, R. Timofte, and L. V. Gool, “Deep expectation of real and apparent age from a single image without facial landmarks,” in IJCV, 2016.
-
- 数据:IMDB-Wiki、Adience、UTKFace、FGNET
- 论文:修改网络结构的SSR-NetIJCAI18 SSR-Net: A Compact Soft Stagewise Regression Network for Age Estimation
-
SSR-Net 年龄和性别
- 数据: IMDB-WIKI dataset、Morph2
- 论文:SSR。速度快
-
DEX 表情年龄性别
相关论文
《Heterogeneous Face Attribute Estimation: A Deep Multi-Task Learning Approach (PAMI- 2017)》
《Deep Attribute Guided Representation for Heterogeneous Face Recognition (IJCAI-18)》
参考blog
额外年龄的数据库汇总
表格中的链接,需要参考这儿
Database | No. of subjects | Database size | Age range (years) |
---|---|---|---|
FG-NET [21] | 82 | 1002 | 0–69 |
MORPH [148] | 13,618 | 55,134 | 27–68 |
Yamaha gender and age (YGA) [12, 68] | 1600 | 8000 | 0–93 |
Waseda human-computer interaction technology [134] | 26,222 | 5500 | 3–85 |
AI & R Asian [150] | 17 | 34 | 22–61 |
Burt’s Caucasian Face database [151] | – | 147 | 20–62 |
Lotus Hill Research Institute (LHI) database [152] | – | 50,000 | 9–89 |
Human and object interaction processing (HOIP) [11] | 300 | 306,600 | 15–64 |
Iranian face database [153] | 616 | 3600 | 2–85 |
Gallagher’s Web-Collected database [4] | – | 28,231 | 0–66 |
Ni’s Web-Collected database [154, 155] | – | 219,892 | 1–80 |
Kyaw’s Web-Collected Database [156] | – | 963 | 3–73 |
BERC database [214] | 95 | 5910 | 3–83 |
3D morphable database [69, 157] | 438 | – | – |