Anonymous A (OP) double-posted this , 3 minutes later[^][v]#1,461,183
Every pixel has 3 subpixels for the 3 primary colors of light: red, green, and blue. The brightness of each subpixel is represented as 8 1’s and 0’s of binary.
Anonymous A (OP) triple-posted this , 2 minutes later, 6 minutes after the original post[^][v]#1,461,184
png images uses lossless compression. PNG uses the DEFLATE algorithm, which basically uses pointers to link back to sequences of bits that were already previously used in the image, making it better for screenshots and images with solid colors.
JPEG uses lossy compression based on discrete cosine transform which basically turns the image into a sum of cosines, and then the smaller details are omitted, resulting in an image that looks mostly the same but uses vastly less binary data.
Mr. Bloody Lemonade !0JqexkOGBc joined in and replied with this , 8 minutes later, 38 minutes after the original post[^][v]#1,461,190
@previous (B)
Another interesting related topic is linear and non linear color spaces. Back when they still used CRT monitors, there’s an interesting coincidence that human brightness perception doesn’t increase linearly as the number of photons increases and CRT monitors also don’t increase in brightness linearly with an increase in voltage. They both coincidently increase with approximately the square root. So that was built into the sRGB color standard. But there are a couple problems with it: it uses 8 bits because the 8 bit byte standard comes from ASCII being designed to handle the Latin character range, but 256 values of intensity actually isn’t enough to cover the full range of colors that humans can see, so more values are used for brighter intensities than darker intensities. This can cause color banding artifacts when images are increased in brightness digitally. And since sRGB is non Linear A naive implementation of brightening an image will lose data, so you have to linearize it, multiply it, then convert it back into sRGB.
Mr. Bloody Lemonade !0JqexkOGBc double-posted this , 2 minutes later, 40 minutes after the original post[^][v]#1,461,191
The difficulty of quantifying all colors humans can see into numbers actually predates digital displays. Attempts to create color spaces that can capture all colors humans can see actually dates back to the 1930s. For example CIE XYZ was an attempt to create a color space that could represent more colors than humans can see, and was linear. Although the sacrifice they made was using 3 primary colors that do not actually exist called X, Y, and Z.
Mr. Bloody Lemonade !0JqexkOGBc joined in and replied with this , 1 hour later, 3 hours after the original post[^][v]#1,461,233
@previous (E)
Oh my mistake, I thought you were being sarcastic and you were trying to convince me that AI can think. In that case if you’re not being sarcastic, I agree with you.
Mr. Bloody Lemonade !0JqexkOGBc double-posted this , 1 minute later, 3 hours after the original post[^][v]#1,461,234
Although FYI convolutional neural networks aren’t generative AI. It’s just a technique used for image classification. It’s how your Google Photos can create categories based on what’s in your pictures. It works and it’s useful and it does what it’s supposed to.
Mr. Bloody Lemonade !0JqexkOGBc triple-posted this , 1 minute later, 3 hours after the original post[^][v]#1,461,235
It’s actually not that difficult to train one yourself. Like not just download an open source model, unlike an LLM it’s at a level of sophistication where it’s possible and feasible that you can actually create one yourself.