วีดีโอ

It's not clear to me how to parallel propagate subdivision into neighboring triangles. Is there a way how to avoid recursive subdivision on half-edges of main mesh triangles?

Hi Jonathan, Just wanted to say thank you for opensourcing your solution. I have been digging into it, and testing it against an obj file with 7 million points, its subdivides it in only a few milliseconds. I have been trying to go directly from obj to subD, and not use the .ccm in-between file. I was wondering if ConcurrentBitField, could be applied directly to the obj data, and in that way accelerating the generation of the cc_Mesh?

Just checking in to see how you are progressing! Anythjng new?

this is neat but wouldnt it be better to just use alembic?

This is amazing. Well done.

Thanks for the awesome video! As far as I understood, you store cbt buffer in GPU memory and looping it in parallel. If the LOD will be big, isn't it really an issue to loop potentially huge amount of nodes and recalculate sum reduction ? And I don't clearly understand from demo, how and where you scale geometry with "PixelsPerEdge" parameter. Does it connected to LOD level and can it be somehow cached ?

This really helped me understand CBT

KSP2 devs just announced they will be replacing their old terain generation algorithm with this one as it their main performance bottleneck at release, and it looks really promising!

Edge hardness coming over from blender would be awesome, but I dont think the default blend importer takes that into Unity

I'd buy a 100 dollar asset store package with this working from blender 👍

Thanks for the presentation! Would there be an advantage in using mesh shaders to compute the tessellated geometry to reduce memory usage?

Hi, very nice presentation. I’m trying to figure out why you need the sum reduction tree to calculate next/prev in the halfedge. Is that because you allow subdivision of arbitrary n-gons? Otherwise, if we only allow up to e.g pentagonal shapes as in the example, wouldn’t it just be sufficient to look at the face-id of the halfedges in the local neighborhood? E.g for 3 in the example around 12:00, we could look at 4 immediately to see that we need to wrap, and then we look at most 5 steps (2 in this case) back to realize we will wrap around to 0. What am I missing? Isn’t it unnecessarily slow to create the sum reduction tree for very large meshes? Say you allow n-gons, then wouldn’t it be sufficient to binary search over the half edges just comparing the face-id until you hit the upper and lower bounds?

Randomly found this while learning to write tesselation for my terrain shader. But what made me smile is randomly hearing one of my favorite ost from one of the most underrated composer out there ♥ cant get tired of the smurfs ost

Outstanding work.

Excellent job done! I hope this will be adapted in industry asap

Hi! I'd love to see the code that generates the SGGX distribution from the mesh. What did you use?

Don't know if that was supposed to be funny but that was the most epic math video I've ever seen lol.

Very nice indeed. Wondering how one would use this with landscape tiles for implementing a landscape system in a game. Can the landscape be split into uniform sized tiles and if so would the resulting tiles be seamless with regards to tessellation/LODs? The goal is to be able to use different shaders per tile.

I still do not understand how the reflection tensor is related to recursive reflection, or is it correct to compare with reflection 1 time?

Just dropping by again to rewatch. Still great! Lol

Anything new Jonathan? I am itching to see your latest work!!!

I dont think I fully understand. Is this used in conjunction with tesselation shaders or is it a complete solution by itself?

You are a research scientist at Unity! How exciting!

Outstanding.

I'd love to know more about this! Can you describe what is happening here?

Awesome work!

Beautiful solution, great explanation. Many thanks : )

I want to learn more about this stuff

Is this less expensive than raytracing simulation? The video only mentions about a GPU implementation, but what would a software implementation be like?

This is awesome! Thank you for the video and documentation.

Excellent demonstration! I'm wondering whether after the mesh is subdivided through your method it could then be further manipulated by some shader process to create surface detail. I'm sort of imagining animated wave patterns on a water surface, or fine wrinkles on a character's skin, or tree bark, for example. Ya, this looks really powerful in adding greater nuance and detail to modeling. Thanks for showcasing your efforts!

Excellent presentation, such an easy to understand explanation of the algorithm! I'm curious in practice in what way the result of the algorithm (in GPU memory) is used to generate geometry.

is this open source? and if so where can I get source code?

U have god level editing years ago also

Co ja tu robię -,-?

Awesome stuff! Just fyi this has been added to CRYENGINE and looks fantastic in practice!

du très bon boulot !

Thank you for the stunning work! I see your implementation uses three additional ifft channels to calculate the jacobian factors. This is significant in terms of performance, so I was wondering whether you also tried to calculate the jacobian through finite diffs of the horizontal displacement (as a post processing step), and if so, how does the resulting quality compare between the two techniques.

Hahahaha very nice choice of music :D All math related videos should have a soundtrack to it :P

For wave displacements do you swap the Gaussian numbers ? I've seen a few papers use this for the height spectrum random Gaussian numbers: (n1 + i*n2) Then for x:z displacement they used this for the spectrum of random Gaussian numbers: (n2 - i*n1) Tessendorf's paper never mentions this for the choppiness of waves but other papers seem to do it without explanation as to why they did it, did you come across this ?

Is this using polygons or something else

Hi Jonathan, which rendering engine did you use to implement this?

Sorry for my layman question, is this a method of tessellation?

Does it mean that for any spherical distribution with analytical sampling it is possible to construct a sampling method for the distribution clipped to an arbitrary spherical cap?

So, the source code of the Unity Engine itself was modified for this paper ? Or what role did Unity Technologies play ? Its not entirely clear.

Will this see the light? I mean in Unity.

What does "this spherical integral is invariant to linear transformations" intuitively mean? I've only recently started reading this blog : mynameismjp.wordpress.com/2016/10/09/sg-series-part-1-a-brief-and-incomplete-history-of-baked-lighting-representations/ so I was able to understand that its difficult to integrate spherical gaussian area lights in real-time with gaussian on probe surface, so we use a cosine-lobe to approximate the area lights. Therefore problem transforms from integrating 2 SG's to simply multiplying an SG with a cosine-lobe. (Please feel free to correct me if I'm wrong). When you say 'this spherical integral is invariant to linear transform" can I understand it as 'you can do as many linear transformations on the area light :- Move it around, deform it(linearly), rotate it...whatever. The irradiance calculation would be affected accordingly(and correctly)" So this basically gives us the power to make indirect lighting of moving area-lights(even though its just linear motion), compared to current method of 'baking' a static-lights scene onto the probe and using those probes interpolate/approximate indirect lighting at runtime. If what I understand is true, then this is definitely interesting development!

We implemented this two years ago, much faster then the method shown (available now to select few developers please e-mail us) th-cam.com/video/hqp7kHPVr58/w-d-xo.html and th-cam.com/video/H5MK3m5vvTs/w-d-xo.html

very nice since this is from adam unity showcase (!?) when will this be available for the community? i really like the volume light effect !

I knew that... You just made up all those words, right? Next party I go to, when someone says something I don't understand, I know all I gotta do is drop "That's almost as complex as Real-Time Polygonal-Light Shading with Linearly Transformed Cosines". yeah, that'll do it. ;D Seriously though, gave you the 63rd "thumbs up". Impressive.