3 ms·

I find the text on the reference decoded image easier to read, despite the colorful speckles surrounding the words. JPEG2PNG doesn't smear the letters a lot, bu

I find the text on the reference decoded image easier to read, despite the colorful speckles surrounding the words. JPEG2PNG doesn't smear the letters a lot, but does appear to slightly deform them. Consider the leading "a" in "add license, readme", whose upper half seems to be closer together after the JPEG2PNG run. However, now I'm wondering whether my impression that the reference decoded text is easier to read is a form of overfitting. I have spent the past 20 years reading low-quality JPEG screenshots on the internet, maybe the perceptual centers of my brain have developed an unblocking algorithm?

stordoff 6y ago
That's a really interesting suggestion. I'm seeing something similar - the jpeg2png output looks cleaner and at first glance more pleasant to look at, but the reference seems both sharper and easier to read. It feels somehow more easier to ignore the more obvious flaws in the reference, and being used to them sounds like a plausible explanation to my layman ear.