The mechanism of how human genes determine next generation’s appearance is well studies by biologists. What extra an artist can do? Can we predict baby’s appearance only by parents’ appearance? Can we generate the baby of any two object, like the daughter of Spongebob and a BMW car? To me, artists+ machine learning is a possible answer. Artists are more familiar with face structure and are more sensitive to what two objects can produce most interesting kid. Machine learning thus responsible for looking for the common underlying law of how appearance is inherited.
In detail, I trained the image data using the feature I constructed which is much less than the computer science guys need. And I proposed several interesting parents combination to see what magic kid they can produce. All these are coded in R using regression. Just one thing to remind, this project still have a further way to go since its performance still has space to promote by enlarging the dataset, improving the feature engineering as well as implementing better machine learning algorithm.
© Copyright Xu Han 2020