arXiv subjects

Jan Novák

Publications and source records attributed to Jan Novák.

At least 19 recordsLinked to original sources

arXiv2026

RGBX-Next: Towards Realistic Generative Rendering from G-Buffers

Diffusion models have achieved impressive results in image, video, and streaming generation. However, compared to traditional 3D rendering, they still lack precise control over the generated output. We believe a viable path forward is to use generative models as learned renderers conditioned on traditionally rendered G-buffers. We introduce RGBX-Next, a unified generative framework for forward and inverse rendering, which allows estimating G-buffers from images, videos, and streams, and rendering realistic images, videos, and streams from G-buffers. Our key contribution is a general recipe for finetuning diffusion transformer (DiT) models into generative forward and inverse renderers. We show that the resulting models achieve high quality in both realistic generative rendering and intrinsic decomposition. We will make all our models publicly available. We believe that the design principles presented in this paper will benefit future research on controllable generative forward and inverse rendering.

cs.CV↗

arXiv2026

New results concerning the curvature of the Universe

One of the fundamental questions we ask in cosmology is what the spatial curvature of the universe is. Experimental observations show that the universe is very close to being spatially flat. However, new results, mainly from measurements by the Planck satellite, indicate that the curvature parameter might be positive. We will, however, examine completely new theoretical results that suggest that we actually live in a closed Universe.

gr-qc↗

arXiv2026

The cosmological inflation inside the cyclic model of the universe

Inflationary cosmology explains the homogeneity and large-scale structure of the universe through a brief epoch of accelerated expansion following the Big Bang. Cyclic cosmologies, in contrast, describe a universe undergoing successive phases of expansion and contraction and can generate primordial perturbations through alternative mechanisms while extending cosmic history beyond a singular beginning. Because these paradigms address different aspects of cosmic evolution, they are often treated separately. Here we explore the possibility that they may instead arise together. In a cosmological scenario involving two scalar fields, one field governs the cyclic evolution of the universe while the other drives an inflationary phase after each cosmological bounce, suggesting that inflation may be a recurring feature of a cyclic universe.

gr-qc↗

arXiv2026

Waveform Modelling for the Laser Interferometer Space Antenna

LISA, the Laser Interferometer Space Antenna, will usher in a new era in gravitational-wave astronomy. As the first anticipated space-based gravitational-wave detector, it will expand our view to the millihertz gravitational-wave sky, where a spectacular variety of interesting new sources abound: from millions of ultra-compact binaries in our Galaxy, to mergers of massive black holes at cosmological distances; from the beginnings of inspirals that will venture into the ground-based detectors' view to the death spiral of compact objects into massive black holes, and many sources in between. Central to realising LISA's discovery potential are waveform models, the theoretical and phenomenological predictions of the pattern of gravitational waves that these sources emit. This white paper is presented on behalf of the Waveform Working Group for the LISA Consortium. It provides a review of the current state of waveform models for LISA sources, and describes the significant challenges that must yet be overcome.

gr-qc↗

arXiv2025

The old problem of the cosmological constant solved?

We formulate an approach to quantum gravity, called the ring paradigm. Gravity is mediated superluminally, and the graviton is described as a phonon on the grid of matter in the Universe. This theory has very interesting applications to cosmology and would ultimately solve the old problem of the cosmological constant. It further gives new impulses to the scalar field theories because the gravitational ring decays to some phantom field. As is obvious, we radically break the Lorentz invariance, which means that some generalization of the Haag-Lopuszanski-Sohnius theorem in quantum field theory is possible.

gr-qc↗

arXiv2025

Generative Detail Enhancement for Physically Based Materials

We present a tool for enhancing the detail of physically based materials using an off-the-shelf diffusion model and inverse rendering. Our goal is to enhance the visual fidelity of materials with detail that is often tedious to author, by adding signs of wear, aging, weathering, etc. As these appearance details are often rooted in real-world processes, we leverage a generative image model trained on a large dataset of natural images with corresponding visuals in context. Starting with a given geometry, UV mapping, and basic appearance, we render multiple views of the object. We use these views, together with an appearance-defining text prompt, to condition a diffusion model. The details it generates are then backpropagated from the enhanced images to the material parameters via inverse differentiable rendering. For inverse rendering to be successful, the generated appearance has to be consistent across all the images. We propose two priors to address the multi-view consistency of the diffusion model. First, we ensure that the initial noise that seeds the diffusion process is itself consistent across views by integrating it from a view-independent UV space. Second, we enforce geometric consistency by biasing the attention mechanism via a projective constraint so that pixels attend strongly to their corresponding pixel locations in other views. Our approach does not require any training or finetuning of the diffusion model, is agnostic of the material model used, and the enhanced material properties, i.e., 2D PBR textures, can be further edited by artists. This project is available at https://generative-detail.github.io.

cs.GR↗

arXiv2024

Real-Time Neural Appearance Models

We present a complete system for real-time rendering of scenes with complex appearance previously reserved for offline use. This is achieved with a combination of algorithmic and system level innovations. Our appearance model utilizes learned hierarchical textures that are interpreted using neural decoders, which produce reflectance values and importance-sampled directions. To best utilize the modeling capacity of the decoders, we equip the decoders with two graphics priors. The first prior -- transformation of directions into learned shading frames -- facilitates accurate reconstruction of mesoscale effects. The second prior -- a microfacet sampling distribution -- allows the neural decoder to perform importance sampling efficiently. The resulting appearance model supports anisotropic sampling and level-of-detail rendering, and allows baking deeply layered material graphs into a compact unified neural representation. By exposing hardware accelerated tensor operations to ray tracing shaders, we show that it is possible to inline and execute the neural decoders efficiently inside a real-time path tracer. We analyze scalability with increasing number of neural materials and propose to improve performance using code optimized for coherent and divergent execution. Our neural material shaders can be over an order of magnitude faster than non-neural layered materials. This opens up the door for using film-quality visuals in real-time applications such as games and live previews.

cs.GR↗

arXiv2023

Wang tiles enable combinatorial design and robot-assisted manufacturing of modular mechanical metamaterials

In this paper, we introduce a novel design paradigm for modular architectured materials that allows for spatially nonuniform designs from a handful of building blocks, which can be robotically assembled for efficient and scalable production. The traditional, design-limiting periodicity in material design is overcome by utilizing Wang tiles to achieve compatibility among building blocks. We illustrate our approach with the design and manufacturing of an L-shaped domain inspired by a scissor-like soft gripper, whose internal module distribution was optimized to achieve an extreme tilt of a tip of the gripper's jaw when the handle part was uniformly compressed. The geometry of individual modules was built on a 3$\times$3 grid of elliptical holes with varying semi-axes ratios and alternating orientations. We optimized the distribution of the modules within the L-shaped domain using an enumeration approach combined with a factorial search strategy. To address the challenge of seamless interface connections in modular manufacturing, we produced the final designs by casting silicone rubber into modular molds automatically assembled by a robotic arm. The predicted performance was validated experimentally using a custom-built, open-hardware test rig, Thymos, supplemented with digital image correlation measurements. Our study demonstrates the potential for enhancing the mechanical performance of architectured materials by incorporating nonuniform modular designs and efficient robot-assisted manufacturing.

cond-mat.mtrl-sci↗

arXiv2023

The methodology for validation of cross sections in quasi monoenergetic neutron field

Dosimetry cross sections are fundamental quantities essential in determination of neutron fluences in points of interest in technologies under heavy radiation load. The most common is their application to Reactor Pressure Vessel aging management, relating to correct estimation of its residual lifetime. The neutron spectrum in various reactor positions has a similar character as the fission spectrum. Due to this fact, the validation of the neutron dosimetry cross sections in reactor fields, or in a very well-known neutron field of 252Cf(s.f.) is sufficient for ensuring their validity in estimations of neutron fluxes in reactor physics. With an increasing field of applications, as in neutron dosimetry in accelerator-based fields or space applications, requests occurred on detailed validation in spectra different from fission ones. This paper presents the testing of a new methodology for the use of quasi monoenergetic neutron fields, where different sensitivity allows validations of the cross-section in different energy regions than in the fission spectrum. The exact shape of the neutron spectrum in the tested fields is determined by stilbene spectrometry and corrected to scattering by calculation, where applicable. The total flux is determined from Ni and Al flux monitors. The evaluated neutron flux in the target set of activation foils is used for calculation of theoretical reaction rate, which is compared with the experimental value determined from gamma activity. This kind of comparison can be understood as validation. It is worth noting, this methodology applied to the IRDFF-II library shows satisfactory agreement for selected reactions.

physics.ins-det↗

arXiv2023

Inverse Global Illumination using a Neural Radiometric Prior

Inverse rendering methods that account for global illumination are becoming more popular, but current methods require evaluating and automatically differentiating millions of path integrals by tracing multiple light bounces, which remains expensive and prone to noise. Instead, this paper proposes a radiometric prior as a simple alternative to building complete path integrals in a traditional differentiable path tracer, while still correctly accounting for global illumination. Inspired by the Neural Radiosity technique, we use a neural network as a radiance function, and we introduce a prior consisting of the norm of the residual of the rendering equation in the inverse rendering loss. We train our radiance network and optimize scene parameters simultaneously using a loss consisting of both a photometric term between renderings and the multi-view input images, and our radiometric prior (the residual term). This residual term enforces a physical constraint on the optimization that ensures that the radiance field accounts for global illumination. We compare our method to a vanilla differentiable path tracer, and more advanced techniques such as Path Replay Backpropagation. Despite the simplicity of our approach, we can recover scene parameters with comparable and in some cases better quality, at considerably lower computation times.

cs.CV↗

arXiv2021

Scalar perturbations of Galileon cosmologies in the mechanical approach within the late Universe

We investigate the Universe at the late stage of its evolution and inside the cell of uniformity 150-370 Mpc. We consider the Universe to be filled at these scales with dust like matter, a minimally coupled Galileon field and radiation. We use the mechanical approach. Therefore, the peculiar velocities of the inhomogeneities as well as the fluctuations of the other perfect fluids can be considered nonrelativistic. Such fluids are said to be coupled because they are concentrated around the inhomogeneities. We investigate the conditions under which the physical Galileon field, i.e. compatible with results of the latest gravitational wave experiments GW150914, GW151226, GW170104, GW170814, GW170817 and GW170608, can become coupled. We know that at the background level coupled scalar fields behave as a two-component perfect fluids: one which mimics a network of frustrated cosmic string and an another one which corresponds to an effective cosmological constant. We found a correction for the Galileon field, which behaves like a matter component.

gr-qc↗

arXiv2021

Real-time Neural Radiance Caching for Path Tracing

We present a real-time neural radiance caching method for path-traced global illumination. Our system is designed to handle fully dynamic scenes, and makes no assumptions about the lighting, geometry, and materials. The data-driven nature of our approach sidesteps many difficulties of caching algorithms, such as locating, interpolating, and updating cache points. Since pretraining neural networks to handle novel, dynamic scenes is a formidable generalization challenge, we do away with pretraining and instead achieve generalization via adaptation, i.e. we opt for training the radiance cache while rendering. We employ self-training to provide low-noise training targets and simulate infinite-bounce transport by merely iterating few-bounce training updates. The updates and cache queries incur a mild overhead -- about 2.6ms on full HD resolution -- thanks to a streaming implementation of the neural network that fully exploits modern hardware. We demonstrate significant noise reduction at the cost of little induced bias, and report state-of-the-art, real-time performance on a number of challenging scenarios.

cs.GR↗

arXiv2020

Neural Control Variates

We propose neural control variates (NCV) for unbiased variance reduction in parametric Monte Carlo integration. So far, the core challenge of applying the method of control variates has been finding a good approximation of the integrand that is cheap to integrate. We show that a set of neural networks can face that challenge: a normalizing flow that approximates the shape of the integrand and another neural network that infers the solution of the integral equation. We also propose to leverage a neural importance sampler to estimate the difference between the original integrand and the learned control variate. To optimize the resulting parametric estimator, we derive a theoretically optimal, variance-minimizing loss function, and propose an alternative, composite loss for stable online training in practice. When applied to light transport simulation, neural control variates are capable of matching the state-of-the-art performance of other unbiased approaches, while providing means to develop more performant, practical solutions. Specifically, we show that the learned light-field approximation is of sufficient quality for high-order bounces, allowing us to omit the error correction and thereby dramatically reduce the noise at the cost of negligible visible bias.

cs.LG↗

arXiv2020

Level-set based design of Wang tiles for modelling complex microstructures

Microstructural geometry plays a critical role in the response of heterogeneous materials. Consequently, methods for generating microstructural samples are increasingly crucial to advanced numerical analyses. We extend Sonon et al.'s unified framework, developed originally for generating particulate and foam-like microstructural geometries of Periodic Unit Cells, to non-periodic microstructural representations based on the formalism of Wang tiles. This formalism has been recently proposed in order to generalize the Periodic Unit Cell approach, enabling a fast synthesis of arbitrarily large, stochastic microstructural samples from a handful of domains with predefined compatibility constraints. However, a robust procedure capable of designing complex, three-dimensional, foam-like and cellular morphologies of Wang tiles has not yet been proposed. This contribution fills the gap by significantly broadening the applicability of the tiling concept. Since the original Sonon et al.'s framework builds on a random sequential addition of particles enhanced with an implicit representation of particle boundaries by the level-set field, we first devise an analysis based on a connectivity graph of a tile set, resolving the question where a particle should be copied when it intersects a tile boundary. Next, we introduce several modifications to the original algorithm that are necessary to ensure microstructural compatibility in the generalized periodicity setting of Wang tiles. Having established a universal procedure for generating tile morphologies, we compare strictly aperiodic and stochastic sets with the same cardinality in terms of reducing the artificial periodicity in reconstructed microstructural samples. We demonstrate the superiority of the vertex-defined tile sets for two-dimensional problems and illustrate the capabilities of the algorithm with two- and three-dimensional examples.

cs.CG↗

arXiv2019

Neural Importance Sampling

We propose to use deep neural networks for generating samples in Monte Carlo integration. Our work is based on non-linear independent components estimation (NICE), which we extend in numerous ways to improve performance and enable its application to integration problems. First, we introduce piecewise-polynomial coupling transforms that greatly increase the modeling power of individual coupling layers. Second, we propose to preprocess the inputs of neural networks using one-blob encoding, which stimulates localization of computation and improves inference. Third, we derive a gradient-descent-based optimization for the KL and the $χ^2$ divergence for the specific application of Monte Carlo integration with unnormalized stochastic estimates of the target distribution. Our approach enables fast and accurate inference and efficient sample generation independently of the dimensionality of the integration domain. We show its benefits on generating natural images and in two applications to light-transport simulation: first, we demonstrate learning of joint path-sampling densities in the primary sample space and importance sampling of multi-dimensional path prefixes thereof. Second, we use our technique to extract conditional directional densities driven by the product of incident illumination and the BSDF in the rendering equation, and we leverage the densities for path guiding. In all applications, our approach yields on-par or higher performance than competing techniques at equal sample count.

cs.LG↗

arXiv2019

Designing modular 3D printed reinforcement of wound composite hollow beams with semidefinite programming

Fueled by their excellent stiffness-to-weight ratio and the availability of mature manufacturing technologies, filament wound carbon fiber reinforced polymers represent ideal materials for thin-walled laminate structures. However, their strong anisotropy reduces structural resistance to wall instabilities under shear and buckling. Increasing laminate thickness degrades weight and structural efficiencies and the application of a dense internal core is often uneconomical and labor-intensive. In this contribution, we introduce a convex linear semidefinite programming formulation for truss topology optimization to design an efficient non-uniform lattice-like internal structure. The internal structure not only reduces the effect of wall instabilities, mirrored in the increase of the fundamental free-vibration eigenfrequency, but also keeps weight low, secures manufacturability using conventional three-dimensional printers, and withstands the loads induced during the production process. We showcase a fully-automatic procedure in detail for the design, prototype manufacturing, and verification of a simply-supported composite machine tool component, including validation with roving hammer tests. The results confirm that the 3D-printed optimized internal structure almost doubles the fundamental free-vibration eigenfrequency, allowing to increase working frequency of the machine tool, even though the ratio between elastic properties of the carbon composite and the ABS polymer used for 3D printing exceeds two orders of magnitude.

physics.app-ph↗

arXiv2018

Wang tiling aided statistical determination of the Representative Volume Element size of random heterogeneous materials

Wang tile based representation of a heterogeneous material facilitates fast synthesis of non-periodic microstructure realizations. In this paper, we apply the tiling approach in numerical homogenization to determine the Representative Volume Element size related to the user-defined significance level and the discrepancy between bounds on the apparent properties. First, the tiling concept is employed to efficiently generate arbitrarily large, statistically consistent realizations of investigated microstructures. Second, benefiting from the regular structure inherent to the tiling concept, the Partition theorem, and statistical sampling, we construct confidence intervals of the apparent properties related to the size of a microstructure specimen. Based on the interval width and the upper and lower bounds on the apparent properties, we adaptively generate additional microstructure realizations in order to arrive at an RVE satisfying the prescribed tolerance. The methodology is illustrated with the homogenization of thermo-mechanical properties of three two-dimensional microstructure models: a microstructure with mono-disperse elliptic inclusions, foam, and sandstone.

physics.app-ph↗

arXiv2017

Deep Scattering: Rendering Atmospheric Clouds with Radiance-Predicting Neural Networks

We present a technique for efficiently synthesizing images of atmospheric clouds using a combination of Monte Carlo integration and neural networks. The intricacies of Lorenz-Mie scattering and the high albedo of cloud-forming aerosols make rendering of clouds---e.g. the characteristic silverlining and the "whiteness" of the inner body---challenging for methods based solely on Monte Carlo integration or diffusion theory. We approach the problem differently. Instead of simulating all light transport during rendering, we pre-learn the spatial and directional distribution of radiant flux from tens of cloud exemplars. To render a new scene, we sample visible points of the cloud and, for each, extract a hierarchical 3D descriptor of the cloud geometry with respect to the shading location and the light source. The descriptor is input to a deep neural network that predicts the radiance function for each shading configuration. We make the key observation that progressively feeding the hierarchical descriptor into the network enhances the network's ability to learn faster and predict with high accuracy while using few coefficients. We also employ a block design with residual connections to further improve performance. A GPU implementation of our method synthesizes images of clouds that are nearly indistinguishable from the reference solution within seconds interactively. Our method thus represents a viable solution for applications such as cloud design and, thanks to its temporal stability, also for high-quality production of animated content.

cs.LG↗