arXiv subjects

Rüdiger Westermann

Publications and source records attributed to Rüdiger Westermann.

At least 19 recordsLinked to original sources

arXiv2026

Bake It Till You Make It: Ultrafast Spatial Texture-Atlas Splatting

Recent extensions of 3D Gaussian Splatting (3DGS) capture fine color details using hash-grid-based appearance parameterization but incur high computational cost during fragment rendering. We introduce a decoupled radiance representation that models low-frequency geometry and view dependent appearance features with 2D surfels while representing high-frequency textures via a view-independent spatial hash grid that is baked into a compact texture atlas. By including sparsity-enhancing optimizations that penalize semi-transparency and per-primitive falloff, our method aggressively prunes insignificant surfels and achieves significantly faster and sparser reconstructions than prior work. Exploiting geometric sparsity and efficient GPU texture mapping, our approach achieves up to a fivefold speedup over 3DGS while preserving state-of-the-art visual fidelity, enabling real-time 4K rendering at 60 FPS on consumer hardware.

cs.CV↗

arXiv2026

Hybrid Latents: Geometry-Appearance-Aware Surfel Splatting

We introduce a hybrid Gaussian-hash-grid radiance representation for reconstructing 2D Gaussian scene models from multi-view images. Similar to NeST splatting, our approach reduces the entanglement between geometry and appearance common in NeRF-based models, but adds per-Gaussian latent features alongside hash-grid features to bias the optimizer toward a separation of low- and high-frequency scene components. This explicit frequency-based decomposition reduces the tendency of high-frequency texture to compensate for geometric errors. Encouraging Gaussians with hard opacity falloffs further strengthens the separation between geometry and appearance, improving both geometry reconstruction and rendering efficiency. Finally, probabilistic pruning combined with a sparsity-inducing BCE opacity loss allows redundant Gaussians to be turned off, yielding a minimal set of Gaussians sufficient to represent the scene. Using both synthetic and real-world datasets, we compare against the state of the art in Gaussian-based novel-view synthesis and demonstrate superior reconstruction fidelity with an order of magnitude fewer primitives.

cs.CV↗

arXiv2025

Inpaint360GS: Efficient Object-Aware 3D Inpainting via Gaussian Splatting for 360° Scenes

Despite recent advances in single-object front-facing inpainting using NeRF and 3D Gaussian Splatting (3DGS), inpainting in complex 360° scenes remains largely underexplored. This is primarily due to three key challenges: (i) identifying target objects in the 3D field of 360° environments, (ii) dealing with severe occlusions in multi-object scenes, which makes it hard to define regions to inpaint, and (iii) maintaining consistent and high-quality appearance across views effectively. To tackle these challenges, we propose Inpaint360GS, a flexible 360° editing framework based on 3DGS that supports multi-object removal and high-fidelity inpainting in 3D space. By distilling 2D segmentation into 3D and leveraging virtual camera views for contextual guidance, our method enables accurate object-level editing and consistent scene completion. We further introduce a new dataset tailored for 360° inpainting, addressing the lack of ground truth object-free scenes. Experiments demonstrate that Inpaint360GS outperforms existing baselines and achieves state-of-the-art performance. Project page: https://dfki-av.github.io/inpaint360gs/

cs.CV↗

arXiv2025

MeSS: City Mesh-Guided Outdoor Scene Generation with Cross-View Consistent Diffusion

Mesh models have become increasingly accessible for numerous cities; however, the lack of realistic textures restricts their application in virtual urban navigation and autonomous driving. To address this, this paper proposes MeSS (Meshbased Scene Synthesis) for generating high-quality, styleconsistent outdoor scenes with city mesh models serving as the geometric prior. While image and video diffusion models can leverage spatial layouts (such as depth maps or HD maps) as control conditions to generate street-level perspective views, they are not directly applicable to 3D scene generation. Video diffusion models excel at synthesizing consistent view sequences that depict scenes but often struggle to adhere to predefined camera paths or align accurately with rendered control videos. In contrast, image diffusion models, though unable to guarantee cross-view visual consistency, can produce more geometry-aligned results when combined with ControlNet. Building on this insight, our approach enhances image diffusion models by improving cross-view consistency. The pipeline comprises three key stages: first, we generate geometrically consistent sparse views using Cascaded Outpainting ControlNets; second, we propagate denser intermediate views via a component dubbed AGInpaint; and third, we globally eliminate visual inconsistencies (e.g., varying exposure) using the GCAlign module. Concurrently with generation, a 3D Gaussian Splatting (3DGS) scene is reconstructed by initializing Gaussian balls on the mesh surface. Our method outperforms existing approaches in both geometric alignment and generation quality. Once synthesized, the scene can be rendered in diverse styles through relighting and style transfer techniques. project page: https://albertchen98.github.io/mess/

cs.CV↗

arXiv2025

SGLDBench: A Benchmark Suite for Stress-Guided Lightweight 3D Designs

We introduce the Stress-Guided Lightweight Design Benchmark (SGLDBench), a comprehensive benchmark suite for applying and evaluating material layout strategies to generate stiff, lightweight designs in 3D domains. SGLDBench provides a seamlessly integrated simulation and analysis framework, including six reference strategies and a scalable multigrid elasticity solver to efficiently execute these strategies and validate the stiffness of their results. This facilitates the systematic analysis and comparison of design strategies based on the mechanical properties they achieve. SGLDBench enables the evaluation of diverse load conditions and, through the tight integration of the solver, supports high-resolution designs and stiffness analysis. Additionally, SGLDBench emphasizes visual analysis to explore the relationship between the geometric structure of a design and the distribution of stresses, offering insights into the specific properties and behaviors of different design strategies. SGLDBench's specific features are highlighted through several experiments, comparing the results of reference strategies with respect to geometric and mechanical properties.

cs.CE↗

arXiv2025

Bayesian Imaging of Interferometric Data from Polarized Electromagnetic Signals

We present an imaging algorithm for polarimetric interferometric data from radio telescopes. It is based on Bayesian statistics and thereby able to provide uncertainties and to incorporate prior information such as positivity of the total emission (Stokes I) or consistency constraints (polarized fraction can only be between 0% and 100%). By comparing our results to the output of the de-facto standard algorithm called CLEAN, we show that these constraints paired with a consistent treatment of measurement uncertainties throughout the algorithm significantly improve image quality. In particular, our method reveals that depolarization canals in CLEAN images do not necessarily indicate a true absence of polarized emission, e.g., after frequency averaging, but can also stem from uncertainty in the polarization direction. This demonstrates that our Bayesian approach can distinguish between true depolarization and mere uncertainty, providing a more informative representation of polarization structures.

astro-ph.IM↗

arXiv2024

APEBench: A Benchmark for Autoregressive Neural Emulators of PDEs

We introduce the Autoregressive PDE Emulator Benchmark (APEBench), a comprehensive benchmark suite to evaluate autoregressive neural emulators for solving partial differential equations. APEBench is based on JAX and provides a seamlessly integrated differentiable simulation framework employing efficient pseudo-spectral methods, enabling 46 distinct PDEs across 1D, 2D, and 3D. Facilitating systematic analysis and comparison of learned emulators, we propose a novel taxonomy for unrolled training and introduce a unique identifier for PDE dynamics that directly relates to the stability criteria of classical numerical methods. APEBench enables the evaluation of diverse neural architectures, and unlike existing benchmarks, its tight integration of the solver enables support for differentiable physics training and neural-hybrid emulators. Moreover, APEBench emphasizes rollout metrics to understand temporal generalization, providing insights into the long-term behavior of emulating PDE dynamics. In several experiments, we highlight the similarities between neural emulators and numerical simulators.

cs.LG↗

arXiv2024

Application of 3D Gaussian Splatting for Cinematic Anatomy on Consumer Class Devices

Interactive photorealistic rendering of 3D anatomy is used in medical education to explain the structure of the human body. It is currently restricted to frontal teaching scenarios, where even with a powerful GPU and high-speed access to a large storage device where the data set is hosted, interactive demonstrations can hardly be achieved. We present the use of novel view synthesis via compressed 3D Gaussian Splatting (3DGS) to overcome this restriction, and to even enable students to perform cinematic anatomy on lightweight and mobile devices. Our proposed pipeline first finds a set of camera poses that captures all potentially seen structures in the data. High-quality images are then generated with path tracing and converted into a compact 3DGS representation, consuming < 70 MB even for data sets of multiple GBs. This allows for real-time photorealistic novel view synthesis that recovers structures up to the voxel resolution and is almost indistinguishable from the path-traced images

cs.GR↗

arXiv2024

Topographic Visualization of Near-surface Temperatures for Improved Lapse Rate Estimation

Numerical model forecasts of near-surface temperatures are prone to error. This is because terrain can exert a strong influence on temperature that is not captured in numerical weather models due to spatial resolution limitations. To account for the terrain height difference between the forecast model and reality, temperatures are commonly corrected using a vertical adjustment based on a fixed lapse rate. This, however, ignores the fact that true lapse rates vary from 1.2 K temperature drop per 100 m of ascent to more than 10 K temperature rise over the same vertical distance. In this work, we develop topographic visualization techniques to assess the resulting uncertainties in near-surface temperatures and reveal relationships between those uncertainties, features in the resolved and unresolved topography, and the temperature distribution in the near-surface atmosphere. Our techniques highlight common limitations of the current lapse rate scheme and hint at their topographic dependencies in the context of the prevailing weather conditions. Together with scientists working in postprocessing and downscaling of numerical model output, we use these findings to develop an improved lapse rate scheme. This model adapts to both the topography and the current weather situation. We examine the quality and physical consistency of the new estimates by comparing them with station observations around the world and by including visual representations of radiation-slope interactions.

physics.ao-ph↗

arXiv2024

fast-resolve: Fast Bayesian Radio Interferometric Imaging

Context: Interferometric imaging is algorithmically and computationally challenging as there is no unique inversion from the measurement data back to the sky maps, and the datasets can be very large. Many imaging methods already exist, but most of them focus either on the accuracy or the computational aspect. Aims: This paper aims to reduce the computational complexity of the Bayesian imaging algorithm resolve, enabling the application of Bayesian imaging for larger datasets. Methods: By combining computational shortcuts of the CLEAN algorithm with the Bayesian imaging algorithm resolve we developed an accurate and fast imaging algorithm which we name fast-resolve. Results: We validate the accuracy of the presented fast-resolve algorithm by comparing it with results from resolve on VLA Cygnus A data. Furthermore, we demonstrate the computational advantages of fast-resolve on a large MeerKAT ESO 137-006 dataset which is computationally out of reach for resolve. Conclusions: The presented algorithm is significantly faster than previous Bayesian imaging algorithms, broadening the applicability of Bayesian interferometric imaging. Specifically for the single channel VLA Cygnus A datasets fast-resolve is about $144$ times faster than resolve. For the MeerKAT dataset with multiple channels the computational speedup of fast-resolve is even larger.

astro-ph.IM↗

arXiv2024

Neural Fields for Interactive Visualization of Statistical Dependencies in 3D Simulation Ensembles

We present the first neural network that has learned to compactly represent and can efficiently reconstruct the statistical dependencies between the values of physical variables at different spatial locations in large 3D simulation ensembles. Going beyond linear dependencies, we consider mutual information as a measure of non-linear dependence. We demonstrate learning and reconstruction with a large weather forecast ensemble comprising 1000 members, each storing multiple physical variables at a 250 x 352 x 20 simulation grid. By circumventing compute-intensive statistical estimators at runtime, we demonstrate significantly reduced memory and computation requirements for reconstructing the major dependence structures. This enables embedding the estimator into a GPU-accelerated direct volume renderer and interactively visualizing all mutual dependencies for a selected domain point.

cs.CV↗

arXiv2024

Design and Optimization of Functionally-graded Triangular Lattices for Multiple Loading Conditions

Aligning lattices based on local stress distribution is crucial for achieving exceptional structural stiffness. However, this aspect has primarily been investigated under a single load condition, where stress in 2D can be described by two orthogonal principal stress directions. In this paper, we introduce a novel approach for designing and optimizing triangular lattice structures to accommodate multiple loading conditions, which means multiple stress fields. Our method comprises two main steps: homogenization-based topology optimization and geometry-based de-homogenization. To ensure the geometric regularity of triangular lattices, we propose a simplified version of the general rank-$3$ laminate and parameterize the design domain using equilateral triangles with unique thickness per edge. During optimization, the thicknesses and orientation of each equilateral triangle are adjusted based on the homogenized properties of triangular lattices. Our numerical findings demonstrate that this proposed simplification results in only a slight decrease in stiffness, while achieving triangular lattice structures with a compelling geometric regularity. In geometry-based de-homogenization, we adopt a field-aligned triangulation approach to generate a globally consistent triangle mesh, with each triangle oriented according to the optimized orientation field. Our approach for handling multiple loading conditions, akin to de-homogenization techniques for single loading conditions, yields highly detailed, optimized, spatially varying lattice structures. The method is computationally efficient, as simulations and optimizations are conducted at a low-resolution discretization of the design domain. Furthermore, since our approach is geometry-based, obtained structures are encoded into a compact geometric format that facilitates downstream operations such as editing and fabrication.

cs.CE↗

arXiv2024

Compressed 3D Gaussian Splatting for Accelerated Novel View Synthesis

Recently, high-fidelity scene reconstruction with an optimized 3D Gaussian splat representation has been introduced for novel view synthesis from sparse image sets. Making such representations suitable for applications like network streaming and rendering on low-power devices requires significantly reduced memory consumption as well as improved rendering efficiency. We propose a compressed 3D Gaussian splat representation that utilizes sensitivity-aware vector clustering with quantization-aware training to compress directional colors and Gaussian parameters. The learned codebooks have low bitrates and achieve a compression rate of up to $31\times$ on real-world scenes with only minimal degradation of visual quality. We demonstrate that the compressed splat representation can be efficiently rendered with hardware rasterization on lightweight GPUs at up to $4\times$ higher framerates than reported via an optimized GPU compute pipeline. Extensive experiments across multiple datasets demonstrate the robustness and rendering speed of the proposed approach.

cs.CV↗

arXiv2024

Postprocessing of Ensemble Weather Forecasts Using Permutation-invariant Neural Networks

Statistical postprocessing is used to translate ensembles of raw numerical weather forecasts into reliable probabilistic forecast distributions. In this study, we examine the use of permutation-invariant neural networks for this task. In contrast to previous approaches, which often operate on ensemble summary statistics and dismiss details of the ensemble distribution, we propose networks that treat forecast ensembles as a set of unordered member forecasts and learn link functions that are by design invariant to permutations of the member ordering. We evaluate the quality of the obtained forecast distributions in terms of calibration and sharpness and compare the models against classical and neural network-based benchmark methods. In case studies addressing the postprocessing of surface temperature and wind gust forecasts, we demonstrate state-of-the-art prediction quality. To deepen the understanding of the learned inference process, we further propose a permutation-based importance analysis for ensemble-valued predictors, which highlights specific aspects of the ensemble forecast that are considered important by the trained postprocessing models. Our results suggest that most of the relevant information is contained in a few ensemble-internal degrees of freedom, which may impact the design of future ensemble forecasting and postprocessing systems.

stat.ML↗

arXiv2024

Adaptive Sampling of 3D Spatial Correlations for Focus+Context Visualization

Visualizing spatial correlations in 3D ensembles is challenging due to the vast amounts of information that need to be conveyed. Memory and time constraints make it unfeasible to pre-compute and store the correlations between all pairs of domain points. We propose the embedding of adaptive correlation sampling into chord diagrams with hierarchical edge bundling to alleviate these constraints. Entities representing spatial regions are arranged along the circular chord layout via a space-filling curve, and Bayesian optimal sampling is used to efficiently estimate the maximum occurring correlation between any two points from different regions. Hierarchical edge bundling reduces visual clutter and emphasizes the major correlation structures. By selecting an edge, the user triggers a focus diagram in which only the two regions connected via this edge are refined and arranged in a specific way in a second chord layout. For visualizing correlations between two different variables, which are not symmetric anymore, we switch to showing a full correlation matrix. This avoids drawing the same edges twice with different correlation values. We introduce GPU implementations of both linear and non-linear correlation measures to further reduce the time that is required to generate the context and focus views, and to even enable the analysis of correlations in a 1000-member ensemble.

cs.GR↗

arXiv2023

Vortices within vortices: hierarchical nature of vortex tubes in turbulence

The JHU turbulence database [1] can be used with a state of the art visualisation tool [2] to generate high quality fluid dynamics videos. In this work we investigate the classical idea that smaller structures in turbulent flows, while engaged in their own internal dynamics, are advected by the larger structures. They are not advected undistorted, however. We see instead that the small scale structures are sheared and twisted by the larger scales. This illuminates the basic mechanisms of the turbulent cascade.

physics.flu-dyn↗

arXiv2023

Transfer Function Optimization for Comparative Volume Rendering

Direct volume rendering is often used to compare different 3D scalar fields. The choice of the transfer function which maps scalar values to color and opacity plays a critical role in this task. We present a technique for the automatic optimization of a transfer function so that rendered images of a second field match as good as possible images of a field that has been rendered with some other transfer function. This enables users to see whether differences in the visualizations can be solely attributed to the choice of transfer function or remain after optimization. We propose and compare two different approaches to solve this problem, a voxel-based solution solving a least squares problem, and an image-based solution using differentiable volume rendering for optimization. We further propose a residual-based visualization to emphasize the differences in information content.

cs.GR↗

arXiv2023

Bayesian radio interferometric imaging with direction-dependent calibration

Context: Radio interferometers measure frequency components of the sky brightness, modulated by the gains of the individual radio antennas. Due to atmospheric turbulence and variations in the operational conditions of the antennas these gains fluctuate. Thereby the gains do not only depend on time but also on the spatial direction on the sky. To recover high quality radio maps an accurate reconstruction of the direction and time-dependent individual antenna gains is required. Aims: This paper aims to improve the reconstruction of radio images, by introducing a novel joint imaging and calibration algorithm including direction-dependent antenna gains. Methods: Building on the \texttt{resolve} framework, we designed a Bayesian imaging and calibration algorithm utilizing the image domain gridding method for numerically efficient application of direction-dependent antenna gains. Furthermore by approximating the posterior probability distribution with variational inference, our algorithm can provide reliable uncertainty maps. Results: We demonstrate the ability of the algorithm to recover high resolution high dynamic range radio maps from VLA data of the radio galaxy Cygnus A. We compare the quality of the recovered images with previous work relying on classically calibrated data. Furthermore we compare with a compressed sensing algorithm also incorporating direction-dependent gains. Conclusions: Including direction-dependent effects in the calibration model significantly improves the dynamic range of the reconstructed images compared to reconstructions from classically calibrated data. Compared to the compressed sensing reconstruction, the resulting sky images have a higher resolution and show fewer artifacts. For utilizing the full potential of radio interferometric data, it is essential to consider the direction dependence of the antenna gains.

astro-ph.IM↗