人脸数据集

MS-Celeb-1M

1. Training dataset, contains 10M images in version 1, is the largest publicly available one in the world 
2. https://www.microsoft.com/en-us/research/project/ms-celeb-1m-challenge-recognizing-one-million-celebrities-real-world/
3. Three kinds of datasets：

•  Full ImageThumbnails data
• cropped face images
•  Aligned face images

LFW

1. LFW数据集主要测试人脸识别的准确率，该数据库从中随机选择了6000对人脸组成了人脸辨识图片对，其中3000对属于同一个人2张人脸照片，3000对属于不同的人每人1张人脸照片。测试过程LFW给出一对照片，询问测试中的系统两张照片是不是同一个人，系统给出“是”或“否”的答案。通过6000对人脸测试结果的系统答案与真实答案的比值可以得到人脸识别准确率。
2. 这个集合被广泛应用于评价 face verification算法的性能
3. http://vis-www.cs.umass.edu/lfw/#download

 MegaFace

1. 包含大约13422个id，87W个样本

2. 部分样本标注错误

3. http://megaface.cs.washington.edu/dataset/download.html

CASIA

1. contains 2,500 color facial images of 500 subjects.
2. http://biometrics.idealtest.org/dbDetailForUser.do?id=9

VGGFace2

• Images are downloaded from Google Image Search and have large variations in pose, age, illumination, ethnicity and profession.

• http://www.robots.ox.ac.uk/~vgg/data/vgg_face2/

DeepGlint-Face

• including MS1M-DeepGlint and Asian-DeepGlint

• https://github.com/deepinsight/insightface/wiki/Dataset-Zoo

• http://trillionpairs.deepglint.com/overview

IJB-B

• 67000 face images, 7000 face videos, and 10000 non-face images.  

• https://www.nist.gov/programs-projects/face-challenges

IJB-C

• 138000 face images, 11000 face videos, and 10000 non-face images.

• https://www.nist.gov/programs-projects/face-challenges

Trillion-Pairs

• http://trillionpairs.deepglint.com/data

• Train dataset called MS-Celeb-1M-v1c with 86,876 ids/3,923,399 aligned images cleaned from MS-Celeb-1M dataset.

YTF（Youtobe Faces DB）

• The data set contains 3,425 videos of 1,595 different people. All the videos were downloaded from YouTube.

• An average of 2.15 videos are available for each subject. The shortest clip duration is 48 frames, the longest clip is 6,070 frames, and the average length of a video clip is 181.3 frame

1.