Face recognition training data: the challenge
The 2- and 3-dimensional automated facial recognition techniques find more and more application possibilities. Aside from their original purpose ¬– which is the identification or verification of persons in the safety and criminal-law domain – numerous fields of application have been added in the economic and research industries. With the help of facial recognition, robots can, for example, recognize human emotions and react accordingly. Online shops selling glasses can offer suitable frames for every user through facial recognition and the clients can virtually try them on. Through automated facial recognition, person-specific styling as well as make-up tips can be given online and new stylings can be visualized.
A high-quality face recognition software should be able to read and track the history of every face effectively. The biometrics of faces, however, differ from person to person. In addition, as soon as a person presents emotional expressions and/or his/her face cannot be seen from the front, the position of his/her mouth, eyes, cheeks, etc. changes. The software should be adapted to each of these features. To develop such a system, it has to be intensively trained with numerous and varied photos of faces. This is where our face recognition training data comes in.
The solution – training data from around the world
Thousands of Clickworkers from various countries and regions create photos of their faces. According to instructions, each Clickworker photographs his/her own face ten times, once head-on and with a neutral facial expression and nine times from different perspectives and with various facial expressions (e.g. anxious, doubtful, angry, laughing, smiling, annoyed, sad, helpless, sulking, or pulling a grimace).
The Clickworkers upload their photos to the platform, stating the respective perspectives and the various facial expressions. Then, the quality management team from clickworker checks all the photographs and their specifications. All correct results are transferred to the customer via an API connection. After a short period of time, he receives high-quality photos in a very efficient manner.
Using these photos, the manufacturer can train the software for facial recognition, optimize it, and specialize the system for different tasks.
More about our service “Training data for AI systems“