First Order Motion Model for Image Animation and Deep Fake Detection: Using Deep Learning

Image animation involves animating an image using a video. In our case, the image contains the face of a person. This face is animated (creation of facial expression) based on the motion of the video frame (may contain a person talking). Current methods involve animating an image with respect to different videos. Our findings show that these methods lack certain features – such as audio capabilities (mimicking the voice of the person in the video). We propose that the chosen image be trained with a variety of videos which contain items of the same classification – say facial features of humans such as eyes, lips, nose etc. and along with the audio output of these items. Self-supervised formulation approach is adapted for the training purpose. Mapping between the motion of the item in the video and the image can be achieved with the help of generator network models. The result is a generation of a model that can mimic any video of the specified classification with the desired audio output.