Face Swap
In this project, we presented three different ways for swapping faces. The first two methods are traditional computer vision methods and the third one is deep-learning method. As part of traditional methods, we presented the detailed explanation of face warping using triangulation and thin plate spline (TPS). For triangulation, we used Delaunay Triangulation method, which is the dual of Voronoi triangulation. In case of deep learning we used the pre-trained MobileNet-V1 to provide landmarks, based on which we implemented face-swap. We used photos of different celebrities and this project members as the data sources and presented the results of face-swap among different inputs.We also considered certain inputs where the results can have artifacts.
Report for this project can be downloaded here.
Instructor: Nitin Sanket