Startup Idea: Save and Sell Porn History

The other day, I stumbled upon Extrapolated Art, which uses machine learning to assess the contents of an image and extrapolate out from there what the image would be like if the canvas had been bigger. The results are often convincing, although sometimes not so much.

This reminds me greatly of Scene Completion using millions of photographs, which purports to "fix" damaged images by analyzing millions of similar photographs, finding one that would fit into the damaged portion of the image, color-matching the image within a given range, and producing a new image.

Why can't we do this with movie frames?

So, my startup idea is simple: Many guys who were college age in the 80s have a strong nostalgia for the porn of that era, most of which is utterly inaccessible due to the loss of original, the utterly crap quality of VHS formats and the slow degradation of VHS tape.

So use the algorithms available to fix it all. Various detection algorithms could identify "damage" in any given frame (including black-box and pixelation censorship). There is a library of zillions of hours of High Definition porn out there. Porn films don't typically have that many scenes and don't typically move the cameras around much; a human would have to vouch for just the very first frame of any given scene, and may have to do some tuning, and then the algorithm would "repair" the movie using each previous frame as a weighted source of quality, turning all those old "classics" into modern, HD versions with balanced colors and crystal clear genitals.

So, all I need is four million dollars, and a couple of other programmers to help me stitch this thing together, and maybe we could tolerate a manager of some kind, and we'll rake in billions, people, billions!.

Touch To Remember

It occurs to me that there are algorithms that can map common objects from photographs, extrapolate the shape of the common object from a library, and render a 3D model of the object.

There is not (yet) a library of millions of 3D scanned human bodies, furniture shapes, and cloth deformations, but the odds are good that we will have one very soon. We could get clothing stores to start scanning clients for custom fits; that would give us a corpus of anonymized human morphologies. Eventually, that library will exist; at which point, the CMU algorithms, image repair, and scene extrapolation will all combine to create perfect 3D worlds. Using Microsoft's "depixelating" algorithm, even the graniest of old movies could be smoothed out, at least at the haptic levels, into something convincing.

Which basically means we're all less than a decade away from being able to get our virtual, but tactile, hands on Christy Canyon and Jeff Stryker.

But for this project, I'll need $12 million.