1. Face database of public figures at Columbia University
Dataset link: http://m6z.cn/5DlIR9
PubFig Dataset is a large face dataset primarily for face recognition and identification, covering 58,797 images of 200 people on the Internet, unlike most existing face datasets, which were taken with the subject completely out of control, so poses, lighting, expressions, scenes, cameras, imaging conditions, and parameters vary greatly from image to image, similar to the LFW dataset created by UMass-Amherst.
The dataset was published by Columbia University in 2009 in the paper "Attribute and Simile Classifiers for Face Verification."
2. CelebA face dataset
Dataset link: http://m6z.cn/60EW0n
CelebFaces Attributes Dataset (CelebA) is a large-scale dataset of face attributes, containing more than 200,000 celebrity images, each with 40 attribute annotations. The images in this dataset cover large pose changes and cluttered backgrounds. CelebA is diverse, numerous, and annotated, including 10,177 identities, 202,599 face images, and 5 landmark locations with 40 binary attribute annotations per image.
The dataset can be used as a training and testing set for the following computer vision tasks: face attribute recognition, face recognition, face detection, landmark (or face part) localization, and face editing and compositing.
3. U.S. Department of Defense Face Bank
Dataset link: http://m6z.cn/5So6DB
In order to promote the research and practical application of face recognition algorithms, the Counterdrug Technology Transfer Program (CTTP) of the U.S. Department of Defense initiated a Face Recognition Technology (FERET) project, which includes a universal face library and common test standards. By 1997, it contained more than 10,000 photos of more than 1,000 people, each with photos of different expressions, lighting, gestures and ages.
4. MTFL face recognition dataset
Dataset link: http://m6z.cn/6fHmaT
The dataset contains 12,995 images of faces annotated with (1) five facial markers, (2) attributes of gender, smile, glasses, and head posture.
5. BioID face dataset
Dataset link: http://m6z.cn/5ZUjyC
This dataset contains 1521 grayscale images with a resolution of 384x286 pixels. Each image comes from a frontal angle of the face of 23 different testers. For ease of comparison, this dataset also contains hand-labeled eye position files corresponding to face images. Images are named in the format "BioID_xxxx.pgm", where xxxx represents the index of the current image (starting at 0). Similarly, files such as "BioID_xxxx.eye" contain the position of the eye in the corresponding image.
6. PersonID face recognition dataset
Dataset link: http://m6z.cn/5So6vR
The face photos selected in this dataset are from two of the more well-known TV series, Buffy the Vampire Hunter and The Big Bang Theory.
7. CMU PIE face library
Dataset link: http://m6z.cn/5vPwfO
CMU PIE Face Library was established in November 2000, it includes 40,000 photos from 68 people, including 13 posture conditions, 43 lighting conditions and 4 kinds of emoticon photos of each person, the existing multi-pose face recognition literature is basically tested on the CMU PIE face library.
8. Youtube video face dataset
Dataset link: http://m6z.cn/6u3P2V
The dataset contains 3,425 videos from 1,595 different people. All videos are downloaded from YouTube. An average of 2.15 videos are available per topic. The shortest clip is 48 frames long, the longest clip is 6070 frames, and the average length of a video clip is 181.3 frames. Under this dataset, the algorithm needs to determine whether the two videos are the same person. There are a number of methods that work on photos that may not be effective/efficient on videos.
9. CASIA face image dataset
Dataset link: http://m6z.cn/5vPwio
Version 5.0 of the CASIA Face Image Database (or CASIA-FaceV5) contains 2,500 color face images for 500 objects. The face image of the CASIA-FaceV5 was captured in one session using a Logitech USB camera. CASIA-FaceV5 volunteers include graduate students, workers, waiters, and more. All face images are 16-bit color BMP files with an image resolution of 640*480. Typical intraclass variations include illumination, posture, expressions, glasses, imaging distance, etc.
10. Caltech Face Database
Dataset link: http://m6z.cn/5So6VP
The dataset contains images of people collected from the web by entering common names in Google Image Search. The coordinates of the center of each front eye, nose, and mouth are provided in the ground fact sheet. This information can be used to align and crop faces or as a basic fact for face detection algorithms. The dataset has 10,524 faces of different resolutions and settings, such as portrait images, crowds, and so on. Faces on the sides or in very low resolution are not marked.