Monterey Bay Aquarium
will byers stan first human second
Alisa U Zemlji Chuda
NASA
Kiana Khansmith
Keni
YOU ARE THE REASON
cherry valley forever
Stranger Things
pixel skylines
Claire Keane
oozey mess
⁂
let's talk about Bridgerton tea, my ask is open
hello vonnie
Cosimo Galluzzi
Xuebing Du
occasionally subtle
Cosmic Funnies
Kaledo Art
seen from France
seen from Canada
seen from United States
seen from United States
seen from Saudi Arabia
seen from Bangladesh
seen from New Zealand
seen from United States
seen from United States
seen from United States
seen from United States
seen from United States
seen from United States
seen from United States
seen from United States
seen from United States
seen from Türkiye
seen from India
seen from Italy
seen from United Kingdom
@rianflo-blog
I'm glad that our ray-traced procedural graphics have won Best VR Scene @ http://vrhackathon.com and hope to see others consider searching for realities beyond explicit geometric modeling. Thanks Chris Birke, Dushan Boshnyak, Jason Fowler and Eriks Strals. Also thanks to Leap Motion. Their VR demos where instantly inspiring.
The Noise Paradox
When modeling with signed distance fields (neat reference page by Inigo Quilez), a popular trick is to add procedural detail via noise functions. For instance, you're starting out with the euclidean distance function to a box and add the scalar result some other function by simple addition or subtraction. What this will do is different from triangle mesh concepts like tesselation or displacement mapping. Simply because you still have a box, but the whole space is distorted and therefore your implicit surface (distance == 0) resembles the complexity of the perturbation. However, there is a limit to how much detail you doing it this way. As the range of values increase, benefits from the distance field for things like ray marching or top-down hierarchical subdivision get lost and things become inaccurate and/or costly.
The noise paradox come from the following: We're adding noise to the space such that we can its detail at the surface but we don't want it further away from the surface at all.
A possible solution I've come up with works for the discrete case (the field is stored in memory / in a texture). It uses a discrete distance transform algorithm which keeps "valid" distances (close to the surface) and propagates them out, overwriting distorted values.
The following is a trivial 2D distance function to a horizontal line plus noise. I've divided the distance value by the length of the gradient in order to get more accurate values close to the surface (messing them up even more in the distant parts).
The algorithm first discards values at a certain range leaving us with only a narrow range.
Now keeping the correct distance values, we can use a vector propagation algorithm (http://webstaff.itn.liu.se/~stegu/aadist/edtaa_preprint.pdf) to restore the euclidean distance field in the far ranges.
While the image is in 2D the technique is trivial to generalize for the 3D case, and it seems a practical way to resolve the noise paradox wherever the iterative filter is feasible.
To noise!
Smooth Volumetric Material Boundaries
One issue with voxel graphics (just like with pixels and triangle rasterization) is aliasing along grid axis planes. Below is a test scene showing the problem with volumetric materials simplified to red, green and blue colors.
Since solid volume boundaries are already neatly shaped the same needs to be true for material compositions. Gladly, a few simple tricks allow for getting pretty decent results (see below). Wouldn't a 3D printer be happy about this too?
Planets born
Looking back a both git log and my history of screenshots I found how the planets started...
The above "planet" came our from a) discovering what OpenGL core profile was and never having written a single shader before and b) wanting a volumetric world. Minecraft world hardly works on a sphere, sure. I've been away from 3D graphics for years and haven't written a single shader yet. But I was totally convinced of how great those APIs must be nowadays.
A screenshot from a week later shows a ray-traced planet. I now remember that I wanted to address the "cubeness" of voxels by treating them only as grid of values rather than a literal cube shape. I also found some translucent volume rendering:
Look & feel is key and the fuzzy voxel look brings up nostalgia of games like Comanche or Outcast, all having been worlds not made out of sharp edges or discrete elements.
I think these "imperfections" will keep allowing our imagination to play, even if we crank up the resolution and detail.
This is my very first post after having an empty account for years and I'd like to debut it with a video of what I've been working on for the past 30 months.
Yes, I'm working on an ambitious virtual world. One that intents to overcome the experiential shortcomings of most video games by rethinking rendering and simulation from the ground up. A world that is alive and in the closest way possible (for a deterministic computer simulation) will resemble a reality, only weakly determined by my own design decisions, not imposing, player driven, AI driven. A nature.
From today, I'm gonna share my experience developing this.