Cody Bennett

@pmndrs, @threejs | Mathematician, Physicist | Graphics, Games, Cats

[Cody Bennett @CodyBennett]

@dangreenheck I have implementations of their later work, and it's difficult since the presentations leave much out on lighting, and most online references are on Horizon: Zero Dawn, including the chapter. I think a physical basis like RDR2 or Unreal would go further.
https://t.co/npXVzrSDYZ

[Cody Bennett @CodyBennett]

Implementing the vertical profile of Nubis/Horizon Forbidden West volumetric clouds with a ground bounce added to the atmosphere (no atmospheric perspective).

Starting to play with the noise composite. First two are the typical Perlin-Worley, last two are with alligator for wisps.

[Cody Bennett @CodyBennett]

@garrettkjohnson https://t.co/mLUMOHFc12

[Cody Bennett @CodyBennett]

@garrettkjohnson They are slower to trace, but faster to update is a big one. I have been slowly working on a conservative error metric for offscreen LOD selection too rather than just clipmaps and hoping shadow casters aren't prominent enough (distance from camera not conservative anymore!).

[Cody Bennett @CodyBennett]

@garrettkjohnson With Radiance Cascades, I found that a BVH2 is better than CWBVH for BLAS, as structured rays are directionally and spatially coherent. I don't see much on this online, as this would be roughly equivalent to optimizing BVH for primary visibility, which would compete with raster.

[Cody Bennett @CodyBennett]

@garrettkjohnson Was just thinking of this after implementing CWBVH + H-PLOC. I precompute CWBVH to BLAS for my clustered ray tracer but don't stream them like I do for my rasterizer (two different DAGs). Awkwardly, I cannot specify up front for HWRT with a streaming scheme.

[Cody Bennett @CodyBennett]

@MrCollison @kvickart @threejs This is not good. Unreal Engine is not permissively licensed, to the point where I will not look at its source code. DDGI is well understood and not complicated to implement, except maybe depth moments. Just do that or even look on to Godot's documentation on SDFGI based on DDGI.

[Cody Bennett @CodyBennett]

Would people be interested in a talk about an SSGI based on this work? Fast and accurate diffuse+specular GI without noise or smudge. Retains high-frequency details in normal maps without TAA for convergence. Was thinking about @threejs/@threejsconf or Khronos/Slang at SIGGRAPH.

[Cody Bennett @CodyBennett]

Some figures from a presentation I gave on Radiance Cascades recently. Covering the theory, intuition, and implementation in 2D and 3D for GI. Plus notes on future research.

[Cody Bennett @CodyBennett]

@alightinastorm @cmiller__ Because you are using my work? This doesn't go for everyone—see the entire thread above.

[Cody Bennett @CodyBennett]

@alightinastorm @cmiller__ This was harmless, as why should I care if you got an expedient result, although perhaps suboptimal? My frustration is with the shrinking knowledge pool as of late with AI and this intellectual honesty. It's not the robots, but people willfully doing this.
https://t.co/fY0bwIKokD

[Cody Bennett @CodyBennett]

@dangreenheck Sad to see and recognize work of myself and others contributed in exchange for nothing but attribution and education to then be convolved and stripped of attribution or knowledge of prior art. What remains is a shrinking circle of knowledge and disenfranchised contributors.

[Cody Bennett @CodyBennett]

@alightinastorm @cmiller__ This is not new at all; people love to skip to results and overlook implementation details. For small things, this is a decent way to learn and can be crafted intentionally for such. But with more intricate things, there is no substitute for understanding what you are working on.

[Cody Bennett @CodyBennett]

@dangreenheck Ideas can only go so far where HW constraints are unforgiving, and so implementation details make/break feasibility. I was hopeful WebGPU would allow more sharing/learning from established engines and practice, but that can only be done with attribution and thus history intact.

[Cody Bennett @CodyBennett]

@dangreenheck Three.js has been a conduit of this lately and I have personally found myself hesitating to contribute any of my own novel research to it or surrounding community. What brought me to the web to openly share and communicate ideas is no longer there, and now I prefer formal forums.

[Cody Bennett @CodyBennett]

@cmiller__ @alightinastorm You can't ask AI to do deduction for you, and that is what is needed for high-level mathematics and physics. It is my hope that the materials for Radiance Cascades aren't too daunting, but please spend some time sitting with it, and of course you can still use AI for impl details

[Cody Bennett @CodyBennett]

@threejs We can use those savings to ship sourcemaps. Helpful for profiling and Lighthouse or anyone using minified bundles. Since three ships its source code (intended or not), much of the size can be removed to just point to the actual JavaScript source.

[Cody Bennett @CodyBennett]

@alightinastorm Clip maps for the shadows look very broken. As does world scale; looks too small. None of the effects you listed are working at a physical scale if visible, especially atmospheric perspective which is a key indication of depth. I would spend some time implementing in isolation.

[Cody Bennett @CodyBennett]

@nagakagachi I was thinking that an adaptive froxel grid would pair well with screen-space probes to capture volumetrics. Can still have a dense probe grid in world space for either to fall back to for accumulation and offscreen radiance.

[Cody Bennett @CodyBennett]

@tonyblu331 The presentation was recorded, but much of it came from an article and research papers I have been working on. Plus live demos I share here. I have difficult personal issues in the way right now, but I am aiming to publish them and get the article and slides together this month.

[Cody Bennett @CodyBennett]

Some figures from a presentation I gave on Radiance Cascades recently. Covering the theory, intuition, and implementation in 2D and 3D for GI. Plus notes on future research.