Annotated Facial Landmarks in the Wild: A Large-scale, Real-world Database for Facial Landmark Localization*
Martin Koestinger, Paul Wohlhart, Peter M. Roth, Horst Bischof
Institute for Computer Graphics and Vision, Graz University of Technology
*The work was supported by the FFG projects MDL (818800) and SECRET (821690) under the Austrian Security Research Programme KIRAS.
This page briefly describes our AFLW databse ...
- a brief description of the AFLW database
- the AFLW license agreement
- instructions to download the data
- an overview of related databases
- frequently asked questions (FAQ)
2012-11-28 New database release! This release includes revised and corrected annotations for roughly 2k faces and also new import scripts for various other face databases (PUT,LFPW,BioId,FaceTracer). See the changelog in the archive for details.
2012-01-10 Updated sqlite3 database file and tools, see the changelog in the archive for details.
2011-12-23 First release of AFLW
Annotated Facial Landmarks in the Wild (AFLW) provides a large-scale collection of annotated face images gathered from Flickr, exhibiting a large variety in appearance (e.g., pose, expression, ethnicity, age, gender) as well as general imaging and environmental conditions. In total about 25k faces are annotated with up to 21 landmarks per image. A short comparison to other important face databases with annotated landmarks is provided here:
|# landmarked imgs.||# landmarks||# subjects||image size||image color||Ref.|
Caltech 10,000 Web Faces
CMU / VASC Frontal
CMU / VASC Profile
|590||6 to 9||-||-||grayscale|||
The motivation for the AFLW database is the need for a large-scale, multi-view, real-world face database with annotated facial features. We gathered the images on Flickr using a wide range of face relevant tags (e.g., face, mugshot, profile face). The downloaded set of images was manually scanned for images containing faces. The key data and most important properties of the database are:
- The database contains about 25k annotated faces in real-world images. Of these faces 59% are tagged as female, 41% are tagged as male (updated); some images contain multiple faces. No rescaling or cropping has been performed. Most of the images are color although some of them gray-scale.
- In total AFLW contains roughly 380k manually annotated facial landmarks of a 21 point markup. The facial landmarks are annotated upon visibility. So no annotation is present if a facial landmark, e.g., left ear lobe, is not visible.
- A wide range of natural face poses is captured The database is not limited to frontal or near frontal faces.
- Additional to the landmark annotation the database provides face rectangles and ellipses. The ellipses are compatible with the FDDB protocol. Further, we include the coarse head pose obtained by fitting a mean 3D face with the POSIT algorithm.
- A rich set of tools to work with the annotations is provided, e.g., a database backend that enables to import other face collections and annotation types. Also a graphical user interface is provided that enables to view and manipulate the annotations.
Due to the nature of the database and the comprehensive annotation we think it is well suited to train and test algorithms for
- facial feature localization
- multi-view face detection
- coarse head pose estimation.
By downloading the database you agree to the following restrictions:
- The AFLW database is available for non-commercial research purposes only.
- You agree not to further copy, publish or distribute any portion of the AFLW database. Except, for internal use at a single site within the same organization it is allowed to make copies of the database.
- All submitted papers or any publicly available text using the AFLW database must cite the following paper:
Annotated Facial Landmarks in the Wild: A Large-scale, Real-world Database for Facial Landmark Localization (pdf, bibtex)
Martin Koestinger, Paul Wohlhart, Peter M. Roth, and Horst Bischof
In First IEEE International Workshop on Benchmarking Facial Image Analysis Technologies, 2011
- The organization represented by you will be listed as users of the AFLW database.
If you agree with the terms of the license agreement contact
Michael Opitz (michael.opitz(at)icg.tugraz.at)
to obtain download instructions.
Please send the email from your official account so we can verify your affiliation and include your
- Position (job title)
- and the intended use
- Unfortunately some annotations still lack in quality. As it is by now not planned to perform benchmarks on AFLW we are are planning to constantly improve the quality of the annotations in the next releases.
- The provided software and code is ment as usage example to AFLW. There is no detailed documentation or plenty of example code.
- We are happy for feedback and discussion, if you have comments just drop as a few lines.
AcknowledgementsThe work was supported by the FFG projects MDL (818800) and SECRET (821690) under the Austrian Security Research Programme KIRAS. We want to thank all people who have been involved in the annotation process, especially, the interns at the institute and the colleagues from the Documentation Center of the National Defense Academy of Austria.
 O. Aran, I. Ari, M. A. Guvensan, H. Haberdar, Z. Kurt, H. I. Turkmen, A. Uyar, and L. Akarun. A database of non-manual signs in turkish sign language. In Proc. Signal Processing and Communications Applications, 2007.
 M. M. Nordstrom, M. Larsen, J. Sierakowski, and M. B. Stegmann. The IMM face database - an annotated dataset of 240 face images. Technical report, Informatics and Mathematical Modelling, Technical University of Denmark, DTU, 2004.