Software Engineer @ NVIDIA Computer Graphics enthusiast from Canton :-) Opinions are my own.
Guangzhou
Joined January 2015
- Tweets2.8K
- Following2K
- Followers875
- Likes4.4K
dydx retweeted
dans un silence assourdissant claude vient de lâcher une dinguerie en physique théorique en réglant 2 problèmes qui saoulaient les chercheurs depuis des décennies avec le coefficient de polyakov, il a capté que les trous noirs peuvent aussi pop par transition ultra rapide dans des tout petits univers et à côté de ça il a posé la limite exacte du fameux modular bootstrapping (je vous laisse chercher) pour faire le grand ménage dans toutes ces théories de gravité quantique un peu bancales qui tournaient en rond
pour vous la faire simple, imaginez que les physiciens galéraient à assembler des pièces de lego cosmiques super complexes sur la façon dont un mini trou noir se forme dans l’univers, bah dites vous queclaude a juste sorti la formule miracle pour tout emboîter d’un coup et jeter toutes les fausses pistes à la poubelle et je trouve ça juste incroyable à observer putain!!!
ce qui rend ce truc complètement fou c’est que c’est le genre d’avancée qui va pour moi irriguer toute la tech de pointe pour les années à venir et ceux qui croient encore que l’ia sert juste à gratter des mails ou à faire des résumés de réunion sont totalement perchés
comprenez qu’on est en train de basculer dans une époque où l’IA devient le moteur absolu de la découverte scientifique
dydx retweeted
Interesting and simple example where the RGB renderer is not correctly rendering.
A spectral renderer, for a more or less realistic paint, will correctly light a green wall from a narrow white light spot pointed to a red wall.
(srgb/agx)
A spotlight hits a red wall and only the wall gets lit. Screen-space GI shoots rays from every pixel, reads the colour they hit on screen, and spills that red onto the floor, boxes and ceiling.
Watch the room 👇
▸ The problem
Direct lighting stops at the first surface. Real light bounces, so a lit red wall tints everything near it. Baked lightmaps capture that, but only for static lights and geometry. Here the light moves every frame.
▸ How it works
The source is a Unity URP renderer feature. It traces screen-space hemisphere rays, not voxel cones, light cascades or baked data:
1. Trace, at half resolution: from each pixel, 2 rays in random cosine-weighted directions around the normal.
2. March each ray against the depth buffer: 4 tiny steps (1.5 cm), 8 medium steps (5 cm), then 40 cm steps, 32 steps in all. A hit counts within 10 cm behind the surface.
3. A hit reads last frame's final image at that point, so light keeps bouncing over frames. Brightness is clamped to cut fireflies.
4. Denoise: reproject the history and average up to 8 frames, then blend 94% history. A 3x3 depth- and normal-aware blur with random spacing follows.
5. Combine: indirect light times the surface colour, added to direct light.
▸ What I hit
Two rays per pixel is mostly noise on its own. Where the history can't be reprojected (a cut, or a surface just uncovered), the source switches to 4 rays with a shorter, coarser march.
Turning away shows the limit. With the red wall off screen, rays that would hit it leave the frame and find nothing, so the red on the boxes fades. The source falls back to reflection probes and sky there; this room has neither, so misses return black.
▸ My build
three.js (r187), WebGL2 fragment passes, 1920 x 1080 with GI at 960 x 540. A G-buffer pass writes colour and normals. I ported the trace, temporal reprojection, spatial blur and combine passes with the source defaults. Two extra denoise passes in the source (a Poisson-disk blur and temporal stabilization) are not ported. Room: 6.4 x 9.2 m, 4 m high, one moving spotlight plus a dim fill light.
▸ Trade-offs
+ Colour bleeding from moving lights with no baking
+ Multiple bounces build up over frames
- Only surfaces on screen can bounce light
- Needs temporal history; fast changes lag and smear
- Thin objects leak through the depth test
- Rendered on a software GPU, so no timings
▸ When to use it
Interiors with moving lights or opening doors, layered over a baked or probe fallback for whatever is off screen.
dydx retweeted
超音波モータ（ピエゾソニックモータ）は電圧を加えると伸縮する圧電セラミックを駆動源とした静音で高トルクなモータです。 一般的な電磁力を利用するモータと違い、コイルや磁石を利用せず摩擦力を利用して作動するため小型・軽量でありながら磁場環境で利用でき、バックラッシュのないモータです。
dydx retweeted
GratNet: A Photorealistic Neural Shader for Diffractive Surfaces
Narayan Kandel, Daljit Singh J.S. Dhillon
arxiv.org/abs/2506.15815
Abstract:
Structural coloration is commonly modeled using wave optics for reliable and photorealistic rendering of natural, quasi-periodic and complex nanostructures. Such models often rely on dense, preliminary or preprocessed data to accurately capture the nuanced variations in diffractive surface reflectances. This heavy data dependency warrants implicit neural representation which has not been addressed comprehensively in the current literature. In this paper, we present a multi-layer perceptron (MLP) based method for data-driven rendering of diffractive surfaces with high accuracy and efficiency. We primarily approach this problem from a data compression perspective to devise a nuanced training and modeling method which is attuned to the domain and range characteristics of diffractive reflectance datasets. Importantly, our approach avoids over-fitting and has robust resampling behavior. Using Peak-Signal-to-Noise (PSNR), Structural Similarity Index Measure (SSIM) and a flipping difference evaluator (FLIP) as evaluation metrics, we demonstrate the high-quality reconstruction of the ground-truth. In comparison to a recent state-of-the-art offline, wave-optical, forward modeling approach, our method reproduces subjectively similar results with significant performance gains. We reduce the memory footprint of the raw datasets by two orders of magnitude in general. Lastly, we depict the working of our method with actual surface renderings.
dydx retweeted
Volumetric Inverse Rendering via Neural Radiative Transfer
Ntumba Elie Nsampi, Adarsh Djeacoumar, Hans-Peter Seidel, Tobias Ritschel, Thomas Leimkühler
arxiv.org/abs/2607.13695
Abstract:
Volumetric inverse rendering seeks to recover the optical properties of participating media from images. Existing approaches either rely on differentiable stochastic light transport simulation, which require substantial algorithmic effort, or use simplified models that fail to capture global illumination. We propose a formulation that reconciles physically complete light transport with general-purpose neural optimization. The optical properties of the medium and the full light field are represented as neural fields and estimated through a joint optimization process. Global illumination is enforced via a residual objective derived from the Radiative Transfer Equation in local differential form, complemented by a volume rendering term along primary viewing rays to mitigate \rev{low-frequency} bias. We demonstrate reconstruction of spatially varying, color-resolved scattering, absorption, and phase function parameters from multi-view images. Beyond reconstruction, the same framework supports learning generative models of participating media with physical optical properties under global illumination.
dydx retweeted
A 2×2 matrix A makes the plane flow by x′ = Ax, and each flow is one of nine kinds: points sink, escape, spiral, circle or settle on a line.
The kind is set by the eigenvalues of A, the roots of λ² − (tr A)λ + det A = 0. det A < 0 gives a saddle. Above det A = (tr A)²/4 they are complex: spirals, or a centre if tr A = 0. Below it, nodes; on it, star and degenerate node.
The stable spiral here has tr A = −1, det A ≈ 2.65, so λ = −0.5 ± 1.55i.
dydx retweeted
Amazing animated globe by @tulparmaps. Have a look at the poorer regions. Lots of progress was achieved in the last 30 years!
dydx retweeted
RTX Remix for Minecraft: Java Edition brings path tracing and DLSS to gaming’s largest modding community!
#RTXOn
dydx retweeted
Try Spectra Render Engine for Blender for Free - Dev Build. #b3d
blenderartists.org/t/spectra…
dydx retweeted
Introducing beignet, a new neural field architecture!
beignet uses a multiresolution fourier frequency pyramid to achieve faster convergence and higher precision when solving PDEs and finding singularities.
Teaser: here is beignet at work fitting an image (super fast)!
dydx retweeted
2024 年科学家们定义了一种新的几何形状“软胞” (Soft Cells)，可以无缝的填充整个空间，我对其中以截角八面体为骨架的形状特别感兴趣：边变成圆弧，面是极小曲面，具备对称美感。
昨天让 ChatGPT Pro 读论文建立数学模型，Codex 接手 CAD 建模，然后 3D 打印机打印出来，心想事成言出法随的好时代啊。
dydx retweeted
Neural Dynamic GI: Random-Access Neural Compression for Temporal Lightmaps in Dynamic Lighting Environments
Jianhui Wu, Jian Zhou, Zhi Zhou, Zhangjin Huang, Chao Li
arxiv.org/abs/2604.12625
Abstract:
High-quality global illumination (GI) in real-time rendering is commonly achieved using precomputed lighting techniques, with lightmap as the standard choice. To support GI for static objects in dynamic lighting environments, multiple lightmaps at different lighting conditions need to be precomputed, which incurs substantial storage and memory overhead.
To overcome this limitation, we propose Neural Dynamic GI (NDGI), a novel compression technique specifically designed for temporal lightmap sets. Our method utilizes multi-dimensional feature maps and lightweight neural networks to integrate the temporal information instead of storing multiple sets explicitly, which significantly reduces the storage size of lightmaps. Additionally, we introduce a block compression (BC) simulation strategy during the training process, which enables BC compression on the final generated feature maps and further improves the compression ratio. To enable efficient real-time decompression, we also integrate a virtual texturing (VT) system with our neural representation.
Compared with prior methods, our approach achieves high-quality dynamic GI while maintaining remarkably low storage and memory requirements, with only modest real-time decompression overhead. To facilitate further research in this direction, we will release our temporal lightmap dataset precomputed in multiple scenes featuring diverse temporal variations.
