/vr/ - Retro Games

>>5259137
old college humor video.

Google was did something similar for their digital zoom on the Pixel. Ultimately they implemented a linear-ish function (the exact function depends on the picture, and is thus not linear, but the individual mappings are) that was trained on a bunch of natural data sets. Ultimately the resulting set of filters looked like the proposed response of simple cells in the lowest levels of the mammalian visual cortex; they kind of looked liked 2D Gaussians multiplied by various spatial frequencies. Something tells me these neural network techniques are homing in on similar filters, although it would be harder to prove.

Interesting stuff. However, does it really make these games better?