[ynnhmhlc] Automatic porn computational challenges

Since human body parts mostly look the same, it should be within the limits of machine learning and computer graphics to take a photograph of a model in a bikini and synthesize an image without the bikini.  Use machine learning on other models who provide training examples: actual photographs with and without clothing but in the same pose.   Also challenging is that some clothing, e.g., push-up bra, moves body parts, so 3D vision and physics simulation is needed to unmove them.  Again, the fact that body parts are similar among people can provide a strong probabilistic prior for 3D reconstruction from a 2D image.

Video also seems with the realm of possibility, though tougher.

But "bubbling", a.k.a. "Mormon porn", seems to work just as well, with your mind, as opposed to the computer's, filling in the details.

Perhaps a combination of the two: computer graphics can easily synthesize patches of featureless skin (e.g., make thin straps disappear), and bubbling to cover the rest.

Bubbling seems to be a bit of an art, for example, wanting to avoid tiny bubbles.  Create a tool for marking which parts should be covered, which exposed, and it automatically bubbles the photograph for you.  While creating a foam of tangent circles is a well studied problem, when we go to "higher order" shapes like ellipses and ovals which have many degrees of freedom, the computational geometry task becomes challenging.