Abstract Great advancements in commodity graphics hardware have favoured processing unit (GPU)‐based volume rendering as the main adopted solution for interactive exploration of rectilinear scalar volumes on platforms. Nevertheless, long data transfer times and GPU memory size limitations are often limiting factors, especially massive, time‐varying or multi‐volume visualization, well networked visualization emerging mobile devices. To address this issue, a variety level‐of‐detail (LOD)...

We address the problem of denoising Monte Carlo renderings by studying existing approaches and proposing a new algorithm that yields state-of-the-art performance on wide range scenes. analyze from theoretical empirical point view, relating strengths limitations their corresponding components with an emphasis production requirements. The observations our analysis instruct design filter offers high-quality results stable performance. A key observation is using auxiliary buffers (normal,...

Perceptually lossless foveated rendering methods exploit human perception by selectively at different quality levels based on eye gaze (at a lower computational cost) while still maintaining the user's of full render. We consider three and propose practical rules thumb for each method to achieve significant performance gains in real-time frameworks. Additionally, we contribute new metric perceptual building HDR-VDP2 that, unlike traditional metrics, considers loss fidelity peripheral vision...

Abstract This paper presents a time‐varying, multi‐layered biophysically‐based model of the optical properties human skin, suitable for simulating appearance changes due to aging. We have identified key aspects that cause such changes, both in terms structure skin and its chromophore concentrations, rely on extensive medical tissue literature accurate data. Our can be expressed biophysical parameters, parameters commonly used graphics rendering (such as spectral absorption scattering...

Large scale and structurally complex volume datasets from high-resolution 3D imaging devices or computational simulations pose a number of technical challenges for interactive visual analysis. In this paper, we present the first integration multiscale representation based on tensor approximation within GPU-accelerated out-of-core multiresolution rendering framework. Specific contributions include (a) hierarchical brick-tensor decomposition approach pre-processing large data, (b) GPU...

This work explores the integration of generative AI models for automatically generating synthetic image-labeled data. Our approach leverages controllable Diffusion Models to generate variations semantically labeled images. Synthetic datasets semantic segmentation struggle represent real-world subtleties, such as different weather conditions or fine details, typically relying on costly simulations and rendering. However, can diverse images using input text prompts guidance images, like masks....

Abstract We present a novel multiresolution compression‐domain GPU volume rendering architecture designed for interactive local and networked exploration of rectilinear scalar volumes on commodity platforms. In our approach, the is decomposed into hierarchy bricks. Each brick further subdivided smaller blocks, which are compactly described by sparse linear combinations prototype blocks stored in an overcomplete dictionary. The dictionary learned, using limited computational memory resources,...

We propose a new adaptive rendering algorithm that enhances the performance of Monte Carlo ray tracing by reducing noise, i.e., variance, while preserving variety high-frequency edges in rendered images through novel prediction based reconstruction. To achieve our goal, we iteratively build multiple, but sparse linear models. Each model has its window, where predicts unknown ground truth image can be generated with an infinite number samples. Our method recursively estimates errors...

Great advancements in commodity graphics hardware have favored GPU-based volume rendering as the main adopted solution for interactive exploration of rectilinear scalar volumes on platforms. Nevertheless, long data transfer times and GPU memory size limitations are often limiting factors, especially massive, time-varying or multi-volume visualization, networked visualization emerging mobile devices. To address this issue, a variety level-of-detail representations compression techniques been...

Variable rate is a requirement for flexible and adaptable image video compression. However, deep compression methods are optimized single fixed rate-distortion tradeoff. While this can be addressed by training multiple models different tradeoffs, the memory requirements increase proportionally to number of models. Scaling bottleneck representation shared autoencoder provide variable with autoencoder. R-D performance using simple mechanism degrades in low bitrates, also shrinks effective...

Unbiased rendering algorithms such as path tracing produce accurate images given a huge number of samples, but in practice, the techniques often leave visually distracting artifacts (i.e., noise) their rendered due to limited time budget. A favored approach for mitigating noise problem is applying learning-based denoisers unbiased noisy and suppressing while preserving image details. However, denoising typically introduce systematic error, i.e., bias, which does not decline rapidly when...

Abstract We present a GPU accelerated volume ray casting system interactively driving multi‐user light field display. The display, driven by single programmable GPU, is based on specially arranged array of projectors and holographic screen provides full horizontal parallax. characteristics the display are exploited to develop specialized rendering technique able provide multiple freely moving naked‐eye viewers illusion seeing manipulating virtual volumetric objects floating in workspace. In...

Direct volume rendering (DVR) using volumetric path tracing (VPT) is a scientific visualization technique that simulates light transport with objects' matter physically-based lighting models. Monte Carlo (MC) often used surface models, yet its application for models difficult due to the complexity of integrating MC light-paths in media none or smooth material boundaries. Moreover, auxiliary geometry-buffers (G-buffers) produced volumes are typically very noisy, failing guide image denoisers...

Compelling virtual reality experiences require high quality imagery as well head motion with six degrees of freedom. Most existing systems limit the viewer (prerecorded fixed position 360 video panoramas), or are limited in realism, e.g. game graphics rendered real-time on low powered devices. We propose a solution for presenting movie to user while still allowing sense presence afforded by free viewpoint motion. By transforming offline content into novel immersive deep media representation,...

Abstract Real-time play recognition and classification algorithms are crucial for automating video production live broadcasts of sporting events. However, current methods relying on human pose estimation deep neural networks introduce high latency commodity hardware, limiting their usability in low-cost real-time applications. We present PlayNet, a novel approach to handball classification. Our method is based Kalman embeddings, new low-dimensional representation game states that enables...

We present an end-to-end system capable of real-time capturing and displaying with full horizontal parallax high-quality 3D video contents on a cluster-driven multiprojector light-field display. The capture component is array low-cost USB cameras connected to single PC. RawM-JPEG data coming fromthe software-synchronized are multicast over Gigabit Ethernet the back-end nodes rendering cluster, where they decompressed rendered. For all-in-focus rendering, view-dependent depth estimated GPU...

Abstract We propose a new real‐time temporal filtering and antialiasing (AA) method for rasterization graphics pipelines. Our is based on Pixel History Linear Models (PHLM), concept modeling the history of pixel shading values over time using linear models. Based PHLM, our can predict per‐pixel variations function between consecutive frames. This combines reprojection with predictions in order to provide temporally coherent shading, even presence very noisy input images. address both spatial...

This article presents a comprehensive dataset of labeled game situations obtained from multiple professional handball matches, which corresponds to the research paper entitled "PlayNet: Real-time Handball Play Classification with Kalman Embeddings and Neural Networks" [1]. The encompasses approximately 11 hours footage five games played in two different arenas, resulting around 1 million data frames. Each frame has been meticulously using seven distinct situation classes (left right attacks,...