How could the checker-board image above be "fixed"? It feels like regardless of what algorithm we use, we're still trying to fit something infinite into a finite number of pixels. Is it possible to light up the right pixels that create the "illusion" of infinity?
How complex is current research on anti-aliasing? I recall reading somewhere that machine learning is also being used to achieve anti-aliasing.
Does ray tracing make anti-aliasing easier? Rather, does the sampling pipeline apply to that way of thinking about graphics, and how?
If (when) our displays reach a high enough pixel density that individual pixels become indistinguishable to the human eye, is (anti-)aliasing still relevant?
tcl1
Is this why videos of confetti on youtube always look horrible compared to the same scene without confetti? Is it just these moire patterns?
djevans
@tcl1 do you watch a lot of confetti videos on youtube? Looking for recommendations.
jkalapos
Aliasing was something I always wondered about when I would play different videogames because there was an option to increase the antialiasing in order to make the graphics appear smoother. It's cool now to know that I was increasing the sampling that my graphics card was doing and also explains why everything got a bit slower
keenan
@lykospirit
1) In some situations you may be able to find an analytical solution. For instance, you might ask, "Can I find a closed form expression for integral of the color of the projected checkerboard over a single pixel?" If so, then you don't even need to bother sampling---just evaluate the integral. In general, though, this is not a realistic approach, as most 'interesting' integrals don't have a closed form and need to be approximated numerically anyway. In short: no, in general you can't get a perfect image. But with enough samples, you can definitely get something very close. :-)
2) Aliasing shows up in a huge number of context (sound, images, motion, video, geometry, ...) so people are of course still asking questions about how to mitigate it. As for the connection between ML and AA, I'm not really sure, though I have heard that the latest NVIDIA cards do some kind of filtering based on learning techniques.
3) Ray tracing gives you more flexibility in where to place samples, but the basic issue of aliasing is still there.
4) Great question. High DPI displays definitely help, but I'm sure I can cook up some crazy scene (like the checkerboard, or better yet, the zone plate) that will still produce an ugly, crawling image on such a display. Not to mention that you still have to worry about temporal aliasing (e.g., due to poor sampling in time).
keenan
@tcl1 Yes, I would also be interested in these confetti videos. :-)
A few things:
How could the checker-board image above be "fixed"? It feels like regardless of what algorithm we use, we're still trying to fit something infinite into a finite number of pixels. Is it possible to light up the right pixels that create the "illusion" of infinity?
How complex is current research on anti-aliasing? I recall reading somewhere that machine learning is also being used to achieve anti-aliasing.
Does ray tracing make anti-aliasing easier? Rather, does the sampling pipeline apply to that way of thinking about graphics, and how?
If (when) our displays reach a high enough pixel density that individual pixels become indistinguishable to the human eye, is (anti-)aliasing still relevant?
Is this why videos of confetti on youtube always look horrible compared to the same scene without confetti? Is it just these moire patterns?
@tcl1 do you watch a lot of confetti videos on youtube? Looking for recommendations.
Aliasing was something I always wondered about when I would play different videogames because there was an option to increase the antialiasing in order to make the graphics appear smoother. It's cool now to know that I was increasing the sampling that my graphics card was doing and also explains why everything got a bit slower
@lykospirit
1) In some situations you may be able to find an analytical solution. For instance, you might ask, "Can I find a closed form expression for integral of the color of the projected checkerboard over a single pixel?" If so, then you don't even need to bother sampling---just evaluate the integral. In general, though, this is not a realistic approach, as most 'interesting' integrals don't have a closed form and need to be approximated numerically anyway. In short: no, in general you can't get a perfect image. But with enough samples, you can definitely get something very close. :-)
2) Aliasing shows up in a huge number of context (sound, images, motion, video, geometry, ...) so people are of course still asking questions about how to mitigate it. As for the connection between ML and AA, I'm not really sure, though I have heard that the latest NVIDIA cards do some kind of filtering based on learning techniques.
3) Ray tracing gives you more flexibility in where to place samples, but the basic issue of aliasing is still there.
4) Great question. High DPI displays definitely help, but I'm sure I can cook up some crazy scene (like the checkerboard, or better yet, the zone plate) that will still produce an ugly, crawling image on such a display. Not to mention that you still have to worry about temporal aliasing (e.g., due to poor sampling in time).
@tcl1 Yes, I would also be interested in these confetti videos. :-)