4 ms·

This is a cool and simple technique, it seems like the sort of thing that should have been discovered 50 years ago. Not only is it more cycle efficient, the lin

by Mizza 3y ago

This is a cool and simple technique, it seems like the sort of thing that should have been discovered 50 years ago. Not only is it more cycle efficient, the lines are more stable in animation, which to me seems like the primary benefit. Amazing this stuff is still being improved.

naillo 3y ago
On that topic there's so many things that are so simple and should have been discovered 50 years. Diffusion feels like it should have been discovered 30 years ago (adding noise to an image and solving for the reverse to generate new images of the same kind).
- david-gpu 3y ago
  > adding noise to an image and solving for the reverse to generate new images of the same kind [should have been discovered 30 years ago] Diffusion networks sre new, but denoising autoencoders work on the principle you outlined and were invented in the eighties. The reason neural nets didn't become popular sooner is because achieving the sorts of results they produce nowadays requires a ton of computing power and a ton of data to train from.
hgsgm 3y ago
"You will have noticed that the slope-based lines above "wiggle" more. This is true, and might be considered one of the drawbacks of the algorithm. However, the fact that BRR lines are more temporally stable can also conceivably be useful in some applications." For nearly all applications, better quality for trivial cost is preferred. The "instability" of classic lies is almost always a benefit -- it provides a very cheap form of motion blur / time-dimension dithering anti-aliasing that hides the ugly pixel aliasing of stable lines. BRR is only better when generating extremely complicated demos, where hyperoptimization matters more than control over the image. For most applications, drawing what the developer/user wants is more important than more complex but arbitrary effects.
jacobgorm 3y ago
Indeed impressive, but compared to when the demo scene fist started, a C64 programmer today has access to all the world’s knowledge via the Internet, and is not constrained by having to host the development environment on the same machine used for testing. Just the fact that you don’t have to reboot the entire setup and possibly even recreate your program every time it crashes, will greatly improve development throughout.

by Mizza 3y ago

naillo 3y ago

On that topic there's so many things that are so simple and should have been discovered 50 years. Diffusion feels like it should have been discovered 30 years ago (adding noise to an image and solving for the reverse to generate new images of the same kind).

david-gpu 3y ago
> adding noise to an image and solving for the reverse to generate new images of the same kind [should have been discovered 30 years ago] Diffusion networks sre new, but denoising autoencoders work on the principle you outlined and were invented in the eighties. The reason neural nets didn't become popular sooner is because achieving the sorts of results they produce nowadays requires a ton of computing power and a ton of data to train from.

hgsgm 3y ago

"You will have noticed that the slope-based lines above "wiggle" more. This is true, and might be considered one of the drawbacks of the algorithm. However, the fact that BRR lines are more temporally stable can also conceivably be useful in some applications." For nearly all applications, better quality for trivial cost is preferred. The "instability" of classic lies is almost always a benefit -- it provides a very cheap form of motion blur / time-dimension dithering anti-aliasing that hides the ugly pixel aliasing of stable lines. BRR is only better when generating extremely complicated demos, where hyperoptimization matters more than control over the image. For most applications, drawing what the developer/user wants is more important than more complex but arbitrary effects.

jacobgorm 3y ago

Indeed impressive, but compared to when the demo scene fist started, a C64 programmer today has access to all the world’s knowledge via the Internet, and is not constrained by having to host the development environment on the same machine used for testing. Just the fact that you don’t have to reboot the entire setup and possibly even recreate your program every time it crashes, will greatly improve development throughout.