This project was centered around warping and morphing different faces into each other.
Part 1 - Defining Correspondences
First, I had to identify keypoints on both my starting and ending images. I hand-selected these, aiming to align to the key features of each person's face.
lebron triangulated
the goat triangulated
Part 2 - Computing the "Mid-Way Face"
This entailed computing the midpoints of all corresponding points, in order to get the triangulation structure of the mid-way face. Then, I warped both Steph and Lebron onto the new mid-way face triangulation. For each face, I did this by iterating through each triangle, inverse warping it onto its corresponding triangle in the mid-way face, and then interpolating to get the colors of each pixel in the new triangle. This successfully moved the colors from the original images onto the mid-way face. From there, I took my two warped basketball players and blended their colors evenly.
original steph
original bron
mid-way face
Part 3 - The Morph Sequence
To create a GIF describing the entire morph from Lebron James to Our Glorious King Wardell Stephen Curry the 2nd, I created 45 morph images, linearly interpolating both warp_frac and dissolve_frac from 0 to 1. The result is below. This guy went from a role player to a top 3 NBA player of all time - give this guy the Most Improved Player award already.
morphing gif (watch a basketball player improve in real time)
Part 4 - The "Mean Face" of a Population
For this exercise, I chose to create the mean face of the FEI dataset, a dataset of Brazilian faces. I calculated the mean face by calculating the mean points for each point of correspondence, getting a mean face from not just two images, but rather an entire set of ~400 (one smiling picture and one neutral picture each for 200 unique faces). I then warped all of the original faces onto this mean face, and then I blended their textures to get the average face. From there, I warped Steph Curry onto this mean face, and then I warped the mean face onto Steph Curry.
Before
After
Warping with Steph Curry
average face of the FEI dataset
the average FEI face warped to the goat's face
the goat's face warped to the average FEI face
Part 5 - Caricatures: Extrapolating from the Mean
For this caricature, I took my Steph Curry and my average FEI face and, instead of interpolating between the two, I extrapolated and tried setting alpha to be a negative value (instead of something between 0 and 1).
alpha=-0.5
alpha=-1.0
alpha=-1.5
alpha=-1.99
Bells & Whistles - Gender Change
For my bells and whistles, I applied a gender change to my favorite basketball player to summon our glorious queen, Wardell Stephanie Curry II. I did this by isolating about 66 female faces from the FEI dataset and calculating their average face. I then applied a morph (and played with the hyperparameters like warp_frac and dissolve_frac) until I found a picture that seemed to resemble both goat-like and female-presenting features.
