Juraj Fulir
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Juraj Fulir has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
- Petra Gospodnetić
Fraunhofer ITWM Germany
4 shared publications - Christoph Garth
RPTU Kaiserslautern Germany
3 shared publications - Runzhou Mao
Fraunhofer ITWM Germany
3 shared publications - Claudia Redenbach
RPTU University Kaiserslautern-Landau
2 shared publications - Christian Freude
Technische Universität Wien (TU Wien), Vienna, Austria
1 shared publication - Christian Jung
Rheinland-Pfälzische Technische Universität Kaiserslautern-Landau (RPTU) Gottlieb-Daimler-Straße 47 67663 Kaiserslautern Germany
1 shared publication
Latest Publications
- Open Access
- 13 November 2026
- Published by Wiley in Computer Graphics Forum
Prompt‐Conditioned Cue Matching for Black‐Box Procedural Material Fitting
Many procedural materials are implemented in legacy or production renderers where differentiable optimization is difficult to integrate, making black‐box parameter fitting still relevant. In this setting, the optimizer receives only a scalar feedback value, so the objective function determines which visual cues are observable to the search. Contrastive Language‐Image Pre‐training (CLIP) image matching is a strong baseline, but its global comparison can still underweight residual cues such as material color, feature scale, stripe spacing, gloss, or cellular structure. We propose Prompt‐Conditioned Cue Matching, a target‐relative CLIP response loss that uses a text prompt as a comparison axis between the target and candidate render. The prompt does not replace the target image; it specifies which cue should be compared more carefully. Because cue‐only optimization is weakly anchored, we combine this term with CLIP image matching in an image‐dominant hybrid loss. Experiments on checker, noise, wave, and Voronoi shaders, together with a real‐capture verification, show that language‐defined cue responses can provide targeted refinement when a visible residual cue remains.
- 13 November 2026
- Published by Wiley in Computer Graphics Forum
Open AccessPrompt‐Conditioned Cue Matching for Black‐Box Procedural Material Fitting
Many procedural materials are implemented in legacy or production renderers where differentiable optimization is difficult to integrate, making black‐box parameter fitting still relevant. In this setting, the optimizer receives only a scalar feedback value, so the objective function determines which visual cues are observable to the search. Contrastive Language‐Image Pre‐training (CLIP) image matching is a strong baseline, but its global comparison can still underweight residual cues such as material color, feature scale, stripe spacing, gloss, or cellular structure. We propose Prompt‐Conditioned Cue Matching, a target‐relative CLIP response loss that uses a text prompt as a comparison axis between the target and candidate render. The prompt does not replace the target image; it specifies which cue should be compared more carefully. Because cue‐only optimization is weakly anchored, we combine this term with CLIP image matching in an image‐dominant hybrid loss. Experiments on checker, noise, wave, and Voronoi shaders, together with a real‐capture verification, show that language‐defined cue responses can provide targeted refinement when a visible residual cue remains.
- Open Access
- 12 August 2026
- Published by Wiley in Computer Graphics Forum
Rendering Synthetic Defects for Learning‐Based Industrial Inspection
Computer vision increasingly uses synthetic data from physically based rendering to supplement limited real‐world datasets. In industrial inspection, defect data is scarce and the rendering pipeline is explicitly controlled, and synthetic defect generation therefore becomes a dataset design problem. In this setting, building a dataset means choosing points in a rendering parameter space: defect shape, material, illumination, viewpoint, and sampling define the training distribution, but their effect on downstream learning is often hard to judge from images alone. We therefore study how these factors change defect features, where their relation to downstream learning is easier to inspect. Our results show that the rendering factors do not matter equally: defect shape, material, illumination, and viewpoint often affect downstream behavior much more than the number of samples per pixel. Synthetic subsets that transfer better downstream tend to stay close to real defect features, cover the observed defect modes, and stay separated from the defect‐free (OK) region. Based on these observations, we build a simple feature‐space screening heuristic for selecting subsets from large candidate pools. The selected subsets often outperform matched random selection for downstream segmentation on real data.
- 12 August 2026
- Published by Wiley in Computer Graphics Forum
Open AccessRendering Synthetic Defects for Learning‐Based Industrial Inspection
Computer vision increasingly uses synthetic data from physically based rendering to supplement limited real‐world datasets. In industrial inspection, defect data is scarce and the rendering pipeline is explicitly controlled, and synthetic defect generation therefore becomes a dataset design problem. In this setting, building a dataset means choosing points in a rendering parameter space: defect shape, material, illumination, viewpoint, and sampling define the training distribution, but their effect on downstream learning is often hard to judge from images alone. We therefore study how these factors change defect features, where their relation to downstream learning is easier to inspect. Our results show that the rendering factors do not matter equally: defect shape, material, illumination, and viewpoint often affect downstream behavior much more than the number of samples per pixel. Synthetic subsets that transfer better downstream tend to stay close to real defect features, cover the observed defect modes, and stay separated from the defect‐free (OK) region. Based on these observations, we build a simple feature‐space screening heuristic for selecting subsets from large candidate pools. The selected subsets often outperform matched random selection for downstream segmentation on real data.
- Open Access
SYNOSIS: Image Synthesis Pipeline for Machine Vision in Metal Surface Inspection
The use of machine learning methods for the development of robust and flexible visual inspection systems has shown promising results. However, their performance is highly dependent on the large amount and diversity of training data, which is difficult to obtain in practice. Recent developments in synthetic dataset generation have seen increasing success in overcoming these problems. However, the prevailing work revolves around the usage of generative models, which suffer from data shortages, hallucinations, and provide limited support for unobserved edge-cases. In this work, we present the first synthetic data generation pipeline that is capable of generating large datasets of physically realistic textures exhibiting sophisticated structured patterns. Our framework is based on procedural texture modelling with interpretable parameters, uniquely allowing us to guarantee precise control over the texture parameters as we generate a high variety of observed and unobserved texture instances. We publish the dual dataset used in this paper, presenting models of sandblasting, parallel, and spiral milling textures, which are commonly present on manufactured metal products. To evaluate the dataset quality, we go beyond final model performance comparison by measuring different image similarities between the real and synthetic domains. This uncovered a trend, indicating these metrics could be used to predict downstream detection performance, which can strongly impact future developments of synthetic data.
Open AccessSYNOSIS: Image Synthesis Pipeline for Machine Vision in Metal Surface Inspection
The use of machine learning methods for the development of robust and flexible visual inspection systems has shown promising results. However, their performance is highly dependent on the large amount and diversity of training data, which is difficult to obtain in practice. Recent developments in synthetic dataset generation have seen increasing success in overcoming these problems. However, the prevailing work revolves around the usage of generative models, which suffer from data shortages, hallucinations, and provide limited support for unobserved edge-cases. In this work, we present the first synthetic data generation pipeline that is capable of generating large datasets of physically realistic textures exhibiting sophisticated structured patterns. Our framework is based on procedural texture modelling with interpretable parameters, uniquely allowing us to guarantee precise control over the texture parameters as we generate a high variety of observed and unobserved texture instances. We publish the dual dataset used in this paper, presenting models of sandblasting, parallel, and spiral milling textures, which are commonly present on manufactured metal products. To evaluate the dataset quality, we go beyond final model performance comparison by measuring different image similarities between the real and synthetic domains. This uncovered a trend, indicating these metrics could be used to predict downstream detection performance, which can strongly impact future developments of synthetic data.
- 12 May 2025
- Published by Springer Nature
- p. 360-377
Sequential PatchCore: Anomaly Detection for Surface Inspection Using Synthetic Impurities
- 12 May 2025
- Published by Springer Nature
- p. 360-377
Sequential PatchCore: Anomaly Detection for Surface Inspection Using Synthetic Impurities
- 24 September 2024
- Published by Springer Nature
- p. 243-256
Simulation of Microstructures and Machine Learning
- 24 September 2024
- Published by Springer Nature
- p. 243-256
Simulation of Microstructures and Machine Learning