arXiv subjects

Dominik Michels

Publications and source records attributed to Dominik Michels.

11 recordsLinked to original sources

arXiv2026

SynGallery: A Synthetic Gallery of Real Paintings for Instance-Level Artwork Recognition

Instance-level artwork recognition requires matching a handheld visitor photograph to a specific work in a large museum collection. This is challenging because painting datasets typically provide clean catalog images for training, while test queries are captured under oblique viewpoints, gallery lighting, reflections, frames, and other scene-level variations. We present SynGallery, a synthetic gallery dataset for artwork retrieval that addresses this gap without collecting additional real photographs. Starting from catalog images of real paintings, we place each artwork into a procedurally generated 3D gallery scene and render it from multiple viewpoints under varied geometric and appearance conditions, while preserving the exact identity of the original work. The resulting dataset contains 24,490 rendered views of 4,898 paintings from the Met benchmark. We show that these synthetic views provide a stronger training signal than the corresponding studio photographs. At the same number of training data points, training only on SynGallery improves art painting recognition from 67.18 to 73.47 GAP$^-$. When added to the full Met training set, SynGallery improves the published benchmark protocol from 35.97 to 38.48 GAP. Ablation experiments show that the gain comes from scene-level view variation rather than photographic realism: reducing the five rendered viewpoints to a single frontal view removes most of the improvement, while simulating capture artifacts such as blur, sensor noise, and image compression consistently reduces performance.

cs.CV↗

arXiv2026

SADGE: Structure and Appearance Domain Gap Estimation of Synthetic and Real Data

We propose SADGE, a quantitative similarity metric that predicts the performance of synthetic image datasets for common computer vision tasks without downstream model training. Estimating whether a synthetic dataset will lead to a model that performs well on real-world data remains a bottleneck in model development. Existing evaluation metrics (e.g., PSNR, FID, CLIP) primarily measure semantic alignment between real and synthetic images (Appearance Similarity Score). Less commonly, structural similarity between images is considered to assess the domain gap (Geometric Similarity Score). However, to the best of our knowledge there exists no studies that evaluate which similarity metric is the best downstream predictor for a given synthetic dataset. In this paper, we show over a wide variety of different synthetic datasets and downstream tasks that neither appearance nor geometry alone can reliably predict downstream performance; rather, it is their non-linear interplay that dictates synthetic data utility. Specifically, we measure how commonly used Appearance and Geometric Similarity metrics computed between synthetic and real images correlate with downstream performance in object detection, semantic segmentation, and pose estimation. Across five public synthetic-to-real benchmark families and 15 dataset-level variants (79k image pairs), SADGE achieves the strongest association with downstream transfer performance under both linear and rank-based criteria, reaching Pearson r=0.88 and Spearman rho=0.77. We compute for each combination of geometry-based methods and appearance-based approaches SADGE scores across all benchmark families. The best configuration is obtained by fusing DINOv3 appearance similarity with MASt3R geometric consistency through a constrained bilinear interaction, outperforming both the strongest geometry-only baseline and the strongest appearance-only baseline .

cs.CV↗

arXiv2026

MaSC: A Masked Similarity Metric for Evaluating Concept-Driven Generation

Evaluating single-concept personalization in text-to-image diffusion requires measuring both concept preservation, which captures identity fidelity to a reference, and prompt following, which captures whether the generated scene matches the prompt. Existing metrics commonly compute these signals using global image or text-image embeddings, such as CLIP-I, DINO, and CLIP-T. We show that such metrics correlate poorly with human perception because they attend to the image as a whole instead of separating the concept subject from the background. We introduce MaSC, a masked similarity metric that uses externally provided foreground concept masks to decompose evaluation into subject-specific concept preservation and background-based prompt following. MaSC computes both scores from frozen SigLIP2 SO400M-NaFlex features: concept preservation is measured by masked max-cosine matching between foreground reference patches and generated-image patches, while prompt following is measured by comparing a background-only pooled image embedding to a subject-stripped prompt embedding. On DreamBench++ human ratings, MaSC achieves Krippendorff alpha = 0.471 for concept preservation, outperforming all tested non-LLM baselines and GPT-4V, and approaching GPT-4o. On ORIDa, a real-photo identity-preservation benchmark across physical environments, MaSC achieves AUC = 0.992, nearly perfectly distinguishing same-subject from cross-subject pairs. Its prompt-following score also outperforms the CLIP-T baseline shipped with DreamBench++. These results show that spatially decomposed aggregation is a strong design principle for evaluating concept-driven generation.

cs.CV↗

arXiv2025

Gaussians on Fire: High-Frequency Reconstruction of Flames

We propose a method to reconstruct dynamic fire in 3D from a limited set of camera views with a Gaussian-based spatiotemporal representation. Capturing and reconstructing fire and its dynamics is highly challenging due to its volatile nature, transparent quality, and multitude of high-frequency features. Despite these challenges, we aim to reconstruct fire from only three views, which consequently requires solving for under-constrained geometry. We solve this by separating the static background from the dynamic fire region by combining dense multi-view stereo images with monocular depth priors. The fire is initialized as a 3D flow field, obtained by fusing per-view dense optical flow projections. To capture the high frequency features of fire, each 3D Gaussian encodes a lifetime and linear velocity to match the dense optical flow. To ensure sub-frame temporal alignment across cameras we employ a custom hardware synchronization pattern -- allowing us to reconstruct fire with affordable commodity hardware. Our quantitative and qualitative validations across numerous reconstruction experiments demonstrate robust performance for diverse and challenging real fire scenarios.

cs.CV↗

arXiv2022

UrbanFlow: Designing Comfortable Outdoor Areas

Design decisions in urban planning have to be made with particular carefulness as the resulting constraints are binding for the whole architectural design that follows. In this context, investigating and optimizing the airflow in urban environments is critical to design comfortable outdoor areas as unwanted effects such as windy areas and the formation of heat pockets have to be avoided. Our UrbanFlow framework enables interactive architectural design allowing for decision making based on simulating urban flow. Compared to real-time fluid flow simulation, enabling interactive architecture design poses an even higher computational efficiency challenge as evaluating a design by simulation usually requires hundreds of time steps. This is addressed based on a highly efficient Eulerian fluid simulator in which we incorporate a unified porosity model which is devised to encode digital urban models containing objects such as buildings and trees. UrbanFlow is equipped with an optimization routine enabling the direct computation of design adaptations improving livability and comfort for given parameterized architectural designs. To ensure convergence of the optimization process, instead of the classical Navier-Stokes equations, the Reynolds-averaged Navier-Stokes equations are solved as this can be done on a relatively coarse grid and allows for the decoupling of the effects of turbulent eddies which are taken into account using a separate turbulence model. As we demonstrate on a real-world example taken from an ongoing architectural competition, this results in a fast convergence of the optimization process which computes a design adaptation avoiding heat pockets as well as uncomfortable windy areas.

cs.GR↗

arXiv2020

Controlling wave-front shape and propagation time with tunable disordered non-Hermitian multilayers

Unique and flexible properties of non-Hermitian photonic systems attract ever-increasing attention via delivering a whole bunch of novel optical effects and allowing for efficient tuning light-matter interactions on nano- and microscales. Together with an increasing demand for the fast and spatially compact methods of light governing, this peculiar approach paves a broad avenue to novel optical applications. Here, unifying the approaches of disordered metamaterials and non-Hermitian photonics, we propose a conceptually new and simple architecture driven by disordered loss-gain multilayers and, therefore, providing a powerful tool to control both the passage time and the wave-front shape of incident light with different switching times. For the first time we show the possibility to switch on and off kink formation by changing the level of disorder in the case of adiabatically raising wave fronts. At the same time, we deliver flexible tuning of the output intensity by using the nonlinear effect of loss and gain saturation. Since the disorder strength in our system can be conveniently controlled with the power of the external pump, our approach can be considered as a basis for different active photonic devices.

physics.optics↗

arXiv2020

Unambiguous scattering matrix for non-Hermitian systems

$\mathcal{PT}$ symmetry is a unique platform for light manipulation and versatile use in unidirectional invisibility, lasing, sensing, etc. Broken and unbroken $\mathcal{PT}$-symmetric states in non-Hermitian open systems are described by scattering matrices. A multilayer structure, as a simplest example of the open system, has no certain definition of the scattering matrix, since the output ports can be permuted. The uncertainty in definition of the exceptional points bordering $\mathcal{PT}$-symmetric and $\mathcal{PT}$-symmetry-broken states poses an important problem, because the exceptional points are indispensable in applications as sensing and mode discrimination. Here we derive the proper scattering matrix from the unambiguous relation between the $\mathcal{PT}$-symmetric Hamiltonian and scattering matrix. We reveal that the exceptional points of the scattering matrix with permuted output ports are not related to the $\mathcal{PT}$ symmetry breaking. Nevertheless, they can be employed for finding a lasing onset as demonstrated in our time-domain calculations and scattering-matrix pole analysis. Our results are important for various applications of the non-Hermitian systems including encircling exceptional points, coherent perfect absorption, $\mathcal{PT}$-symmetric plasmonics, etc.

physics.optics↗

arXiv2018

Symbolic-Numeric Integration of the Dynamical Cosserat Equations

We devise a symbolic-numeric approach to the integration of the dynamical part of the Cosserat equations, a system of nonlinear partial differential equations describing the mechanical behavior of slender structures, like fibers and rods. This is based on our previous results on the construction of a closed form general solution to the kinematic part of the Cosserat system. Our approach combines methods of numerical exponential integration and symbolic integration of the intermediate system of nonlinear ordinary differential equations describing the dynamics of one of the arbitrary vector-functions in the general solution of the kinematic part in terms of the module of the twist vector-function. We present an experimental comparison with the well-established generalized alpha-method illustrating the computational efficiency of our approach for problems in structural mechanics.

math.AP↗

arXiv2017

Algorithmic Verification of Linearizability for Ordinary Differential Equations

For a nonlinear ordinary differential equation solved with respect to the highest order derivative and rational in the other derivatives and in the independent variable, we devise two algorithms to check if the equation can be reduced to a linear one by a point transformation of the dependent and independent variables. The first algorithm is based on a construction of the Lie point symmetry algebra and on the computation of its derived algebra. The second algorithm exploits the differential Thomas decomposition and allows not only to test the linearizability, but also to generate a system of nonlinear partial differential equations that determines the point transformation and the coefficients of the linearized equation. Both algorithms have been implemented in Maple and their application is illustrated using several examples.

math.CA↗

arXiv2016

On the General Analytical Solution of the Kinematic Cosserat Equations

Based on a Lie symmetry analysis, we construct a closed form solution to the kinematic part of the (partial differential) Cosserat equations describing the mechanical behavior of elastic rods. The solution depends on two arbitrary analytical vector functions and is analytical everywhere except a certain domain of the independent variables in which one of the arbitrary vector functions satisfies a simple explicitly given algebraic relation. As our main theoretical result, in addition to the construction of the solution, we proof its generality. Based on this observation, a hybrid semi-analytical solver for highly viscous two-way coupled fluid-rod problems is developed which allows for the interactive high-fidelity simulations of flagellated microswimmers as a result of a substantial reduction of the numerical stiffness.

math.AP↗

arXiv2016

Local Decoders for the 2D and 4D Toric Code

We analyze the performance of decoders for the 2D and 4D toric code which are local by construction. The 2D decoder is a cellular automaton decoder formulated by Harrington which explicitly has a finite speed of communication and computation. For a model of independent $X$ and $Z$ errors and faulty syndrome measurements with identical probability we report a threshold of $0.133\%$ for this Harrington decoder. We implement a decoder for the 4D toric code which is based on a decoder by Hastings arXiv:1312.2546 . Incorporating a method for handling faulty syndromes we estimate a threshold of $1.59\%$ for the same noise model as in the 2D case. We compare the performance of this decoder with a decoder based on a 4D version of Toom's cellular automaton rule as well as the decoding method suggested by Dennis et al. arXiv:quant-ph/0110143 .

quant-ph↗