Combinations of Images?

Had a strange idea. Let's say we have a lofi camera that creates a 3 x 3 image with 9 pixels in total. Each pixel can be 1 of 16 colors. The possible images that can be taken with this camera should be 16^9. It should be possible to programmatically create all of those images instead of having to take them. If you were able to scale this up to a higher definition, wouldn't you be able to find meaningful images within this enormous image dataset? Akin to monkeys writing Shakespeare. Not really sure what I'm going for at the moment but wondering if it's possible with today's technology. Would I be able to recreate the images with a resolution of 800*600? The dataset size would be mind boggling but some sort of filtering may be possible with machine learning.

>meaningful images Define "meaningful" >The dataset size would be mind boggling but some sort of filtering may be possible with machine learning. How do you train your model? How would a machine learning model (i.e. a classifier, a generator or a decision-maker) even perform "filtering" over a fuсkhuge 16^480000-images large dataset? And what are you even filtering for? >>oh but anon, generating an image is like picking one from the set of all images so it's like if you filtered through all of them to get one :​^​)​)) I look forward to such refined, sophisticated perspectives.

>>6967 Hey you never know, inside one of those is Taylor Swift marrying the Pope, I get that this is impractical, being that there is no way (like you said) to efficiently train an ai to search for "meaningful" images, but it's still fun to speculate

https://www.reddit.com/r/allthepictures http://pixelmonkeys.org/ And on the musical side... http://allthemusic.info/

>>6966 >The dataset size would be mind boggling That's putting it mildly. >3x3 For a single image, that's 16^(3 * 3) combinations of images to run through, or 68,719,480,000. In terms of storage, that's 9 pixel positions using 4 bits to represent colours, or 36 bits per image. The total comes to 309,237,600,000 bytes. This is pretty big, but nothing crazy. Any stock desktop or laptop made in the past five years ought to handle this. >800*600 That's 16^(800 * 600) combinations of images to run through. I have yet to figure out a way to represent that number in regular scientific notation. I can get to 16^255 (an image of 15x17 pixels) before everything goes nuts and my program just calls the result "infinity". This is roughly 1.1235582092889474 * 10^307 combinations. To put that number in perspective, there's an estimated 3.28 x 10^80 protons, neutrons, and electrons in the observable universe. You'd need something on the order of 10^227 universes worth of matter, with every single electron and quark used with perfect efficiency to represent 1 bit per particle, just to store every possible 16-colour image in a 15x17 grid. If we assume the average desktop computer with a half-dozen 8TB drives, and if you limit image resolution to a total of 10 pixels (arranged however you like, say 2x5), you'll be able to squeeze a complete dataset into a regular desktop machine. Go to 11 pixels total and you'd need ten desktop machines with a half-dozen 8TB drives. Exponentiation is fun! :) This is also a great way to demonstrate the effectiveness of simply using longer keys in cryptography. Adding a few bits is a trivial way to make a brute-force attack impossible in practice, if the underlying algorithm is sound.

>>6968 >inside one of those is Taylor Swift marrying the Pope You will also find all the winning numbers for every lottery in the indeterminate future, along with detailed instruction on how to make the best possible investments in the stock market. Inside one of those is literally anything you can visually imagine (and more) as long as you can translate it to 16 colors and fit it within 800x600 without losing essential details.