|Téma:||Facial Landmark Annotation Tool|
|Vedoucí:||Ing. Jan Čech Ph.D.|
|Popis:||Facial landmarks are characteristic points on a human face that can be repeatedly detected, eg. eye corners, tip of the nose. Techniques to detect facial landmarks based on machine learning exist nowadays. Their success often rely on a quality (and quantity) of training data (images + positions of the landmarks).
The proposed project will be on developing an efficient tool for semi-automatic annotation. It means a human annotator will be clicking into the image to mark the landmarks, while an algorithm will update a statistical model to propose most likely position of the remaining landmarks to minimize the human effort while maximize the annotation accuracy.
Another aspect of the project is to do the annotation and statistically evaluate accuracy of human annotators and propose a mechanism to detect possibly wrong annotation or an unreliable annotator.
The task will be to:
(1) Develop a GUI application for manual landmark annotation
(2) Include a response from an automatic system (The algorithm will be provided)
(3) Do the annotation for a number of exemplar images
(4) Evaluate annotation statistics
|Max.number of students:||4|