Animating The Mona Lisa Effect With Deep Learning Using Tensorflow.js

The fusion of deep learning with inanimate objects around us has been the tech industry’s recent trend. Image animation is a technique that simulates movement into a still image. In one of her projects, Emily Xie programmed an intuitive digital portrait of Mona Lisa. 

It uses deep learning, image processing techniques, and Tensorflow.js. 

  1. Users need to generate a sequence of images of Mona Lisa’s face looking in different directions. 
  2. The model requires continuously selecting and displaying a unique frame of the eyes gaze based on the viewer’s location as detected by the webcam.  

The system used the First Order Motion Model(FOMM). The algorithm is composed of two modules: 

  • To extract the motion of the user to detect its user. 
  • To generate the gaze of eyes from the pool of images based on the viewer’s location. 

Techniques used: TensorflowJS (BlazeFace), First Order Motion Model, Python, Javascript.

You can check out the project here



Consulting Intern: Grounded and solution--oriented Computer Engineering student with a wide variety of learning experiences. Passionate about learning new technologies and implementing it at the same time.