Results

Part 2: Computing the "Mid-way Face"

Then, to find the affine matrix A, I essentially solved a system of equations using np.linalg.lstsq in my computeAffine(tri1, tri2) function. The a, b, c, d, e, f values were then used with a row of [0, 0, 1] to create the transformation matrix T with homogenous coordinates.

The inverse warp function was then created by generating a mask using polygon. I also used a RegularGridInterpolator in order to interpolate color from neighbors in the case of a pixel coming from "between pixels."

Part 3: The Morph Sequence

The morph is similar, except my mean points is recomputed at each step of iteration using the warp_fraction weight to adjust the shape transformation, and the dissolve_fraction to control the values blending.

I iterated over 0-1 at 45 even steps in order to create the gif transition frames, and create the gif. The transition can be seen below:

Part 4: The "Mean face" of a population

I computed the average face of the Danes, which was a bit tricky because I needed to create a special parser in order to identify the correspondence points of each face. With this info, I averaged the position of all of the correspondence points for each image and then morphed each face into the average face points. I then computed the mean of all of these faces to compute the "average" face.

Here are some examples of people in the dataset with their face morphed to the average shape.

I also have examples of my face being morphed into the average face shape, and the average face being morphed into my face shape.

Part 5. Caricatures: Extrapolating from the mean

I extrapolated from the Danish population mean in order to create a caricature of my face, and tried it with alphas 1 and -0.7.

Bells & Whistles

For my bells & whistles, I focused on changing my ethnicity to the average Dane similar to the previous part, but with more adjustments including shape, appearance, and both.