Grandparent doesn't understand information theory. True superresolution is impossible. ML hallucination is a guess, not actual information recovery. Recovering the information from nowhere breaks the First Law of Thermodynamics. If grandparent can do it, he/she will be immediately awarded the Shannon Award, the Turing Award, and the Nobel Prize for Physics.
True superresolution is impossible, but a heck of a lot of resolution is hidden in video, without resorting to guesses and hallucination.
Tiny camera shakes on a static scene gives away more information about that scene, it's effectively multisampling of the static scene. (If I have to hunch it, any regular video can "easily" be upscaled 50% without resorting to interpolation or hallucination.)
Our wetware image processing does the same - look at a movie shot at the regular 24fps where people walk around. Their faces look normal. But pause any given frame, and it's likely a blur. (But our wetware image processing likely does hallucination too, so it's maybe not a fair comparison.)
It's not temporal interpolation. It's using data from different frame(s) to fill in the current frame. It's not interpolation at all. It's using a different accurate data source to augment another.
Super resolution can and does work in some circumstances.
By introducing a new source of information (the memory of the NN) it can reconstruct things it has seen before, and generalise this to new data.
In some cases this means hallucinations, true. But in other times (eg text where the NN has seen the font) it is reconstructing what that font is from memory.
But the thing is, in that case the information contained in the images was actually much less than what we are meant to make believe.
So if we are reconstructing letters from a known font we essentially are extracting 8 bits of information from the image. I'm pretty certain that if you distort the image to an SNR equivalent of below 8 bits you will not be able to extract the information.
Lossy image compression creates artifacts, which are in a way of form of falsely reconstructed information - information which wasn't there in the original image. Lossless compression algorithms work by reducing redundancy, but don't create information where it wasn't there (thus being very different from super-resolution algorithms).
Not if it’s written text and you are selecting between 26 different letters. It’s a probabilistic reconstruction, but that’s very different to a hallucination.
