|Topic:||Facial Time Lapse Video|
|Supervisor:||Ing. Jan Čech Ph.D.|
|Description:||A facial time lapse video documents an age progression of a person. The video might be cool to visualize e.g., child growing.
The input would be an unorganized collection of images (a family photo album, automatically crawled images of politicians or other celebrities). A target person should be present in most of the images, but may appear on an image together with other persons or may not be present at all (on small number of images).
The algorithm would then process the input set: (1) Detects faces, (2) Finds the target person, estimate the acquisition time of the image (in case EXIF data is unavailable or is apparently wrong). (3) Detects facial landmark points to find the correspondences to register the faces. (4) Morph the temporarily subsequent facial images to generate a smooth transition.
The output will be a video presented to the user.
Your task will be to prepare a web-service, when a user upload his/her image collection, the algorithm is executed on the server to output the video available for download.
We will provide you all necessary elements to script the final algorithm: face detector, identity descriptor (convolutional net), age estimator (convolutional net), facial landmark detector. You will need to implement the registration and morphing, which is not difficult.
You will get in touch with the rapidly evolving deep convolutional network technology. The project workload can be naturally distributed among several team members. Impressive results are expected.
|Max.number of students:||0|