DataThis page provides links to the following data sets
Unfiltered faces for gender and age classification
DescriptionIn order to facilitate the study of age and gender recognition, we provide a data set and benchmark of face photos. The data included in this collection is intended to be as true as possible to the challenges of real-world imaging conditions. In particular, it attempts to capture all the variations in appearance, noise, pose, lighting and more, that can be expected of images taken without careful preparation or posing.
The sources of the images included in our set are Flickr albums, assembled by automatic upload from iPhone5 (or later) smart-phone devices, and released by their authors to the general public under the Creative Commons (CC) license.
This data set was used in the paper Age and Gender Estimation of Unfiltered Faces (See our publications page for more information). The paper describes the process of collecting the data set and provides additional information on the test protocols used with it.
Statistics and infoTotal number of photos: 26,580
Total number of subjects: 2,284
Number of age groups / labels: 8 (0-2, 4-6, 8-13, 15-20, 25-32, 38-43, 48-53, 60-)
Gender labels: Yes
In the wild: Yes
Subject labels: Yes
FilesFiles for this benchmark are available under the subdirectory named:
The directory contains the following files:
- faces.tar.gz (936M) - Face images, cropped
- aligned.tar.gz (1.9G) - Face images, cropped and aligned using our 2D, in-plain alignment tool
- fold_0_data.txt - fold_4_data.txt - text files with indices to the five-fold cross validation tests using all faces
- fold_frontal_0_data.txt - fold_frontal_4_data.txt - same as above, but using only faces in approximately frontal pose
- README.txt provides details on these text files
- LICENSE.txt provides license information for the face images in this collection.
DownloadThe data is stored on our FTP server: agas.openu.ac.il
Alternatively, direct access to the download directory is available through:
To get access to the data, please provide the details below. We ask for your details so we can keep in touch in case we find any need for a critical update or in case we would organize a dedicated workshop, etc. When done, you will immediately be able to see the FTP user name and password.
Important note regarding downloads: If the direct link above fails, please try the FTP server. For various technical reasons, if you do so, we strongly advised that you use a dedicated FTP client to access the FTP server (rather than using the web browser). Two free clients are FileZilla and WinSCP.
ErrataDec. 15th, 2014: We have corrected a number of wrong gender labels. The new split (text) files have been uploaded to the FTP. The previous split files are still available for reference and backwards comparability. We are most grateful for Jordi Mansanet Sandin for noticing and reporting these label errors.
Nov. 17th, 2014: Thanks to reports from some of our benchmark users, we have located and corrected a few errors in the split files. These errors were introduced when compiling the data for download; they do not affect the scores reported in the paper. We have replaced the old split files with the new, corrected versions on the FTP server. If you have downloaded the split files before the 17th of Nov. 2014, we ask that you please download the corrected files.