Wenzel Jakob
@wjakob.bsky.social
Associate professor leading EPFL's Realistic Graphics Lab. My research involves inverse graphics, material appearance modeling and physically based rendering
You can dump the PTX intermediate representation (see the documentation), but figuring out the calling convention of the kernel for your own use will be tricky. The system is not designed to be used in this way.
August 22, 2025 at 10:36 AM
You can dump the PTX intermediate representation (see the documentation), but figuring out the calling convention of the kernel for your own use will be tricky. The system is not designed to be used in this way.
Just write your solver in plain CUDA. How hard can it be? 😛
August 18, 2025 at 7:33 PM
Just write your solver in plain CUDA. How hard can it be? 😛
This approach is restricted to software that only needs the CUDA driver. If your project uses cuSolver, you will likely need to have a dependency on the CUDA python package that ships this library on PyPI (similar to PyTorch et al.)
August 18, 2025 at 7:09 PM
This approach is restricted to software that only needs the CUDA driver. If your project uses cuSolver, you will likely need to have a dependency on the CUDA python package that ships this library on PyPI (similar to PyTorch et al.)
Wasn’t that something.. Flocke (German for “flake”) says hi!
August 15, 2025 at 4:45 AM
Wasn’t that something.. Flocke (German for “flake”) says hi!
If you are fitting a NeRF and you want a surface out at the end, you should probably be using the idea in this paper.
August 12, 2025 at 6:43 AM
If you are fitting a NeRF and you want a surface out at the end, you should probably be using the idea in this paper.
Given the focus on performance, I would suggest to switch from pybind11 to nanobind. Should just be a tiny change 😇
August 11, 2025 at 3:05 PM
Given the focus on performance, I would suggest to switch from pybind11 to nanobind. Should just be a tiny change 😇
For the paper and data, please check out the project page: mokumeproject.github.io
Mokume
mokumeproject.github.io
August 8, 2025 at 11:53 AM
For the paper and data, please check out the project page: mokumeproject.github.io
This video explains the process more detail: www.youtube.com/watch?v=H6N-...
The Mokume Dataset and Inverse Modeling of Solid Wood Textures (Technical Paper, SIGGRAPH 2025)
YouTube video by Maria Larsson
www.youtube.com
August 8, 2025 at 11:53 AM
This video explains the process more detail: www.youtube.com/watch?v=H6N-...
The Mokume project is a massive collaborative effort led by Maria Larsson at the University of Tokyo (w/Hodaka Yamaguchi, Ehsan Pajouheshgar, I-Chao Shen, Kenji Tojo, Chia-Ming Chang, Lars Hansson, Olof Broman, Takashi Ijiri, Ariel Shamir, and Takeo Igarashi).
August 8, 2025 at 11:53 AM
The Mokume project is a massive collaborative effort led by Maria Larsson at the University of Tokyo (w/Hodaka Yamaguchi, Ehsan Pajouheshgar, I-Chao Shen, Kenji Tojo, Chia-Ming Chang, Lars Hansson, Olof Broman, Takashi Ijiri, Ariel Shamir, and Takeo Igarashi).
To reconstruct their interior, we: 1️⃣Localize annual rings on cube faces 2️⃣ Optimize a procedural growth field that assigns an age to every 3D point (when that wood formed during the tree's life) 3️⃣ Synthesize detailed textures via procedural model or a neural cellular automaton
August 8, 2025 at 11:53 AM
To reconstruct their interior, we: 1️⃣Localize annual rings on cube faces 2️⃣ Optimize a procedural growth field that assigns an age to every 3D point (when that wood formed during the tree's life) 3️⃣ Synthesize detailed textures via procedural model or a neural cellular automaton
The Mokume dataset consists of 190 physical wood cubes from 17 species, each documented with:
- High-res photos of all 6 faces
- Annual ring annotations
- Photos of slanted cuts for validation
- CT scans revealing the true interior structure (for future use)
- High-res photos of all 6 faces
- Annual ring annotations
- Photos of slanted cuts for validation
- CT scans revealing the true interior structure (for future use)
August 8, 2025 at 11:53 AM
The Mokume dataset consists of 190 physical wood cubes from 17 species, each documented with:
- High-res photos of all 6 faces
- Annual ring annotations
- Photos of slanted cuts for validation
- CT scans revealing the true interior structure (for future use)
- High-res photos of all 6 faces
- Annual ring annotations
- Photos of slanted cuts for validation
- CT scans revealing the true interior structure (for future use)
Wood textures are everywhere in graphics, but realistic texturing requires knowing what wood looks like throughout its volume, not just on the surfaces.
The patterns depend on tree species, growth conditions, and where and how the wood was cut from the tree.
The patterns depend on tree species, growth conditions, and where and how the wood was cut from the tree.
August 8, 2025 at 11:53 AM
Wood textures are everywhere in graphics, but realistic texturing requires knowing what wood looks like throughout its volume, not just on the surfaces.
The patterns depend on tree species, growth conditions, and where and how the wood was cut from the tree.
The patterns depend on tree species, growth conditions, and where and how the wood was cut from the tree.
The reason is that the "volume" of this paper is always rendered as a surface (without alpha blending) during the optimization. Think of it as an end-to-end optimization that accounts for meshing, without actually meshing the object at each step.
August 8, 2025 at 8:01 AM
The reason is that the "volume" of this paper is always rendered as a surface (without alpha blending) during the optimization. Think of it as an end-to-end optimization that accounts for meshing, without actually meshing the object at each step.
To get a triangle mesh out at the end, you will still need a meshing step (e.g. marching cubes). The key difference is that NeRF requires addtl. optimization & heuristics to create a volume that will ultimately produce a high quality surface. With this new method, it just works.
August 8, 2025 at 8:01 AM
To get a triangle mesh out at the end, you will still need a meshing step (e.g. marching cubes). The key difference is that NeRF requires addtl. optimization & heuristics to create a volume that will ultimately produce a high quality surface. With this new method, it just works.
Check out our paper for more details at rgl.epfl.ch/publications...
August 7, 2025 at 12:21 PM
Check out our paper for more details at rgl.epfl.ch/publications...
This is a joint work with @ziyizh.bsky.social, @njroussel.bsky.social, Thomas Müller, @tizian.bsky.social, @merlin.ninja, and Fabrice Rousselle.
August 7, 2025 at 12:21 PM
This is a joint work with @ziyizh.bsky.social, @njroussel.bsky.social, Thomas Müller, @tizian.bsky.social, @merlin.ninja, and Fabrice Rousselle.
Our method minimizes the expected loss, whereas NeRF optimizes the loss of the expectation.
It generalizes deterministic surface evolution methods (e.g., NvDiffrec) and elegantly handles discontinuities. Future applications include physically based rendering and tomography.
It generalizes deterministic surface evolution methods (e.g., NvDiffrec) and elegantly handles discontinuities. Future applications include physically based rendering and tomography.
August 7, 2025 at 12:21 PM
Our method minimizes the expected loss, whereas NeRF optimizes the loss of the expectation.
It generalizes deterministic surface evolution methods (e.g., NvDiffrec) and elegantly handles discontinuities. Future applications include physically based rendering and tomography.
It generalizes deterministic surface evolution methods (e.g., NvDiffrec) and elegantly handles discontinuities. Future applications include physically based rendering and tomography.
Instead of blending colors along rays and supervising the resulting images, we project the training images into the scene to supervise the radiance field.
Each point along a ray is treated as a surface candidate, independently optimized to match that ray's reference color.
Each point along a ray is treated as a surface candidate, independently optimized to match that ray's reference color.
August 7, 2025 at 12:21 PM
Instead of blending colors along rays and supervising the resulting images, we project the training images into the scene to supervise the radiance field.
Each point along a ray is treated as a surface candidate, independently optimized to match that ray's reference color.
Each point along a ray is treated as a surface candidate, independently optimized to match that ray's reference color.
By changing just a few lines of code, we can adapt existing NeRF frameworks for surface reconstruction.
This patch shows the necessary changes to Instant NGP, which was originally designed for volume reconstruction.
This patch shows the necessary changes to Instant NGP, which was originally designed for volume reconstruction.
August 7, 2025 at 12:21 PM
By changing just a few lines of code, we can adapt existing NeRF frameworks for surface reconstruction.
This patch shows the necessary changes to Instant NGP, which was originally designed for volume reconstruction.
This patch shows the necessary changes to Instant NGP, which was originally designed for volume reconstruction.
It also adds support for function freezing so that the process of rendering a scene can be captured and cheaply replayed. See mitsuba.readthedocs.io/en/latest/re... for details.
Release notes - Mitsuba 3
mitsuba.readthedocs.io
August 7, 2025 at 11:15 AM
It also adds support for function freezing so that the process of rendering a scene can be captured and cheaply replayed. See mitsuba.readthedocs.io/en/latest/re... for details.