Tomas Akenine-Möller

Real-Time Rendering progress! (Note that both image and paper quality is much better in the 2nd print) @pointinpolygon @renderwonk @SebHillaire @kenpex @miwanicki

Our upcoming HPG 2025 paper ”Collaborative Texture Filtering” is up! Research done with @_pandersson, Matt Pharr, and @BartWronsk. research.nvidia.com/labs/rtr/publi…

@pixelmager @NOTimothyLottes Here it is: github.com/erich666/jgt-c…

@NOTimothyLottes @inversepixel source for the above paper disappeared with akpeters, any chance you would be able to recover it? :)

For the love of octahedrons: fileadmin.cs.lth.se/graphics/resea… Mapping this well to the GPU follows ...

I'm excited to share our latest paper (work with Matt Pharr and @inversepixel ), "Improved Stochastic Texture Filtering Through Sample Reuse" accepted to I3D'2025. arxiv.org/abs/2504.05562 research.nvidia.com/labs/rtr/publi… 1/

@jendrikillner Thanks for the kind words about our book! 🙏🏻

I often get asked about my book recommendations for graphics programming and game development. I updated my old list and made it easier to find on my website. With Black Friday/Cyber Monday, many are on sale right now (at least in Canada/US) jendrikillner.com/books/

@mwesterdahl76 @ChristerEricson I’m not on X very often any longer. Not even Mastodon. So no advice for you. Would be fun to catch up some day, though. Are you based in Stockholm?

Who should I follow on da Blue skyeee? @ChristerEricson are you there? @inversepixel long time no see?

FLIP v1.4 is out! It makes the Python implementation about 20-47x faster than before, and significantly more convenient to use. Code available at github.com/NVlabs/flip. Details in 🧵.

Single header FLIP.h, for both CPU-only and CUDA, has just been released: github.com/NVlabs/flip

A Korean translation of "Real-Time Rendering" will be released July 31. See realtimerendering.com/blog/korean-ed… for bookstores carrying it.

We finally get to share our neural materials research. We created custom highly-realistic reference materials, we invented new neural models, and we built new graphics systems. End result is these neural materials run in a real-time path tracer. research.nvidia.com/labs/rtr/neura…

NVIDIA is thrilled to share our SIGGRAPH 2023 papers. It is our best SIGGRAPH yet, with ~20 papers authored in partnership with over a dozen universities. Generative AI, 3D content creation, neural rendering, neural physics, and more... blogs.nvidia.com/blog/2023/05/0…

I'm super excited to finally share our work (with @karthikvnathan @marcosalvi @inversepixel @_pandersson @aaronlefohn ), accepted to Siggraph 2023: research.nvidia.com/labs/rtr/neura… Neural Texture Compression.

@phyronnaz Sorry for the late reply. For each n_ijk, we compute the analytic normal, i.e., Eq 9-11. This is 3 bilinear interpolation times 8 corners. And then a trilinear interpolation on top of that. @SoderlundHerman

@inversepixel Very interesting! If I understood correctly, to compute a normal you do 9 trilinear interpolations? 8 to compute the normals from the 8 voxels intersecting a dual voxel, and a final one to merge them?

Lookit! Lookit!

@knarkowicz We did not look carefully into that, unfortunately.

@inversepixel Inspiring! I was curious if you investigated how voxel interesection precision affects results. I imagine larger error would make ST faster, at some point allow to replace DDA+NR/ST with pure ST in 7^3 bricks etc. But I'm curious what's the crossing point and visual tradeoff here

Improved material & light separation in triangular 3D model extraction from images. Paper, video & code: nvlabs.github.io/nvdiffrecmc/ #NeurIPS2022, @nvidia

The #NVIDIA ADA GPU Architecture Whitepaper is available: New RT cores with Opacity Micromaps + Displaced Micro-Meshes, 2x Ray-Triangle intersect throughput, Shader Execution Reordering, 96MB L2 cache, 2x Optical Flow accel. perf, improved Tensor Cores... images.nvidia.com/aem-dam/Soluti…

@marcosalvi @BrianKaris We had it as a TODO in one of our files, but did not get around to it :-(

@marcosalvi @BrianKaris Marco is right: we did not have occupancy bits. That would have helped further.

@BrianKaris No, we did not do that... that's clever! We could do that even for SVS, i.e., shrink the AABB around each surface inside a single voxel (but we have not tried that either).

@inversepixel Also did you test with AABBs shrunk to the surface voxels or were they always full brick sized?