Rendering shadows remains a fundamental problem in computer graphics, with existing approaches balancing performance against visual quality. Traditional shadow mapping techniques are efficient but suffer from aliasing artifacts due to its discretized nature. On the other hand, ray tracing can produce high quality shadow without discretization but is often too demanding for consumer hardware.

This thesis proposes a neural shadow representation that models a continuous mapping from light rays to occluder depth. Instead of relying on discretized buffers, we represent each object's shadow with a fully connected neural network. Given a ray origin and direction, the network predicts the depth at which the ray intersects geometry. This allows for depth tests similar to shadow mapping while supporting continuous input. The model is trained using ray-traced supervision with a dead-zone loss function that encourages the model to output depth corresponding to a position inside the occluder's geometry. It does not rely on discretized shadow maps and supports fully dynamic scenes, point lights, and directional lights.

Our results show that this method can achieve visual quality comparable to medium-resolution shadow maps while eliminating aliasing. The proposed method also has a unique ability to overfit to restricted light configurations. In scenes such as outdoor scenes where light movement is limited, the model's capacity is concentrated on relevant directions, allowing it to capture finer geometric details. While limitations remain regarding inference speed and high-frequency geometry, the proposed method demonstrates unique strengths that open up new trade-offs, particularly in scenarios with partially restrained lighting configurations and large scale but low-frequency geometry such as landscapes. ... Read more

The Neurophysiological Biomarker Toolbox (NBT) visualises biomarkers on two-dimensional cortical parcellations, where rendering cost limits use at clinical scale. Currently, NBT fits each region's boundary independently, leaving gaps and overlaps between adjacent regions that simplification only worsens. This work applies a shared-arc representation that is gap- and overlap-free by construction. Because boundaries are shared, simplification acts on both sides of the boundary at once, so gaps and overlaps cannot form during simplification. Two failure modes remain: a boundary can cross another, or a region can collapse to a degenerate polygon. Five established simplification algorithms are evaluated on the representation across six atlases and across three axes: geometric fidelity, topological consistency, and rendering cost.

The evaluation shows that the algorithms differ little on fidelity. The axis separating them is topological consistency, specifically whether they allow collapse. This matters for cost because in the evaluated algorithms the dominant predictor of rendering time is the number of faces, not the number of vertices. The algorithms that provide the strongest topological guarantees therefore deliver the least cost relief, since they collapse the fewest faces. On this representation, simplification stays free of gaps and overlaps at a small, bounded fidelity cost. ... Read more

Electroencephalography (EEG) often relies on topographic scalp maps (topomaps) to visualize how biomarkers vary across the scalp. These visualizations can be generated many times during biomarker research, making higher efficiency pipelines important for reduced server load and latency. This research compares a raster based topomap rendering pipeline with a vector/SVG based pipeline in terms of server-side and client-side latency, and memory usage for EEG biomarker visualization.

A benchmark was implemented to measure the server-side cost of both pipelines over 100 iterations across 3 montage sizes and 5 interpolation resolutions using randomized EEG-like data. Within the tests, the vector pipeline was found to have server-side speedups between 2.88× and 5.29× while reducing memory usage between 1.24× and 1.42× dependent on the resolution. Stage-level analysis of the two pipelines showed that the raster pipeline was dominated by the rendering stage, whereas the vector pipeline distributed its cost mainly across interpolation and figure construction. However, improvements in server rendering came at the cost of client-side performance, with latency ratios ranging from 1.03× to 0.50× and memory-reduction ratios ranging from 0.96× to 0.14× depending on the resolution.

The results indicate that direct vector SVG rendering can significantly reduce server-side latency and memory usage. However, since the vector approach shifts part of the rendering work to the browser, the client impact remains an important concern. Overall, the proposed vector-based pipeline is a promising approach for modernizing EEG biomarker topomap visualization, but its benefits are dependent on its usage and the consideration of the client-side performance costs. ... Read more

Electroencephalography (EEG) biomarker pipelines are usually assumed to be limited by statistical computation, but much of their cost is data movement: copying, reshaping, and indexing arrays through high-level Python abstractions. We profile the data-reshaping (reduce) stage of the Neurophysiological Biomarker Toolbox, a Python EEG-analysis framework, and find it performs no arithmetic at all: its cost is an eager deep copy of the per-subject container, which also doubles the stage’s peak memory by allocating a full duplicate of the data.
We evaluate three memory-management strategies (zero-copy array views, layout pinning, and lazy materialisation) against an unmodified baseline, verifying that every variant reproduces the baseline’s statistical outputs exactly. Zero-copy views remove the duplicate, cutting reduce-stage peak memory from gigabytes to near zero; this lowers worst-case (tail) latency and, under the parallel load of a cohort sweep, lifts throughput by up to 4.4× by keeping concurrent workers out of swap. Layout pinning and lazy materialisation act only when subjects have repeated sessions, where lazy materialisation cuts reduce-stage peak memory by two orders of magnitude.
The reduce stage is thus effectively eliminated as a cost. The end-to-end speedup is a more modest 1.3×, bounded not by the optimisation but by a separate, arithmetic-bound statistics step that lies outside this paper’s scope and which we flag as the natural next target. The practical recommendation is to eliminate eager deep copies first: a small change that removes the memory doubling and, under parallel load, keeps a cohort sweep out of swap. ... Read more

Electroencephalography is a widely used non-invasive technique used for measuring brain activity. While relatively cheap, the large volumes of data it produces can make analysis and visualization challenging. NBT Cloud aims to address these challenges by bringing EEG analysis to the cloud and creating an ecosystem for standardized, real-time analysis. Because the application must visualize large data in real time, downsampling the signals is necessary. Modern EEG analysis toolkits, however, focus on downsampling while keeping the data's statistical properties, which introduces computational overhead. This paper investigates whether other downsampling methods, such as Min-Max or Largest Triangle Three Buckets, can achieve better visual fidelity at lower runtime. The results show that simpler candidate algorithms better preserve the visual characteristics of EEG signals while achieving lower runtimes. Among the evaluated algorithms, Min-Max offers the best trade-off between visual similarity and performance, making it the most suitable choice for the use case of NBT Cloud. ... Read more

This paper investigates memory-aware optimization of mass-univariate EEG statistical inference in the Neurophysiological Biomarker Toolbox. A vectorized Python implementation and a native Rust backend are evaluated as optimized alternatives to the existing NumPy/SciPy-based statistical testing pipeline. The optimized implementations reorganize EEG biomarker data for cohort-based access, improving support for cache locality, SIMD execution, and parallel processing. Synthetic benchmarks show speedups of up to 452.3x for the vectorized Python implementation and up to 486.1x for the Rust backend. The optimized implementations also substantially reduce sensitivity to increasing biomarker counts, resulting in much weaker runtime growth across the measured benchmark space. Profiling shows increased SIMD density and CPU utilization, while cache behaviour improves only modestly. These results suggest that the primary limitation is not the statistical operation itself, but the overhead introduced by how the workload is structured and executed. Much of the available speedup can therefore be achieved by expressing the computation as larger batched and vectorized operations.
... Read more

Inferring social intention in everyday settings is challenging because the same observable behavior can support multiple plausible interpretations. This issue is pronounced in restaurants, where roles and norms structure interaction but do not uniquely determine what is socially “meant.” The research question addressed is: How can scenarios in a restaurant setting be created that allow investigation into how humans and intelligent systems construct multiple plausible narratives of social intention?A literature-grounded scenario design method is presented that separates observable cues from inferred psychological meanings and situation classes, and frames interactions using external scripts (norms, roles, scenes) and internal script variants (observer-dependent interpretations). Two short scenarios with three controlled variations each are specified to modulate openendedness through cue completeness, norm clarity, and perspective. It is concluded that openendedness can be designed systematically by controlling observability and introducing norm tensions and perspective differences that keep multiple narratives simultaneously plausible. ... Read more

One of AR's promising developments is the delivery of critical information at just the right time.
Light field displays (LFDs) stand out as a potential foundation for AR. Several AR applications use simple HUDs, primarily projecting 2D elements (e.i. outlines, text, glyphs) to deliver critical information. Despite this, most approaches still employ complex 3D rendering techniques to display the content.
Our approach, by contrast, leverages the content's 2D nature to achieve lower sampling times, more efficient memory representation and can be extended to support 2D animated content.
We build on top of work in light field displays, allowing us to maintain correct focus cues and stereoscopy.
More specifically, we report a 100-fold improvement in sampling times, and a minor improvement in rendering time. ... Read more

To incentivize solar panel adaptation, a multitude of solutions are actively being researched. Among these solutions are colored solar cells, using coatings or filters to open the door for more architectural expression. In order to help architects explore these possibilities, it is necessary to enable flexible, fast, and realistic visualization. This work explores the modeling of color-coated photovoltaic cells using a physically based bidirectional reflectance distribution function (BRDF), with a focus on the pyramidically textured structures embedded inside many of these cells. The BRDF is analytically derived, modeling light interactions as recursively specular multiple-scattering. The model is parametric, characterizing the surface using its pyramid density and pyramid slant angle, making it generally applicable to homogeneous pyramidic surfaces with uniform pyramid heights. Evaluation indicates the model closely approximates the behavior of generated references at steep viewing angles. Clear avenues of improvement, including corrections at shallow angles, are discussed within the context of future work. ... Read more

Lens flares occur when a bright light source induces light to travel through an optical lens system via unintended paths, reaching the sensor at an undesired location. Although arising from the imperfection of lens systems, flares are widely used in the visual entertainment industry for artistic purposes and to increase perceived brightness.

Research on physically accurate lens flare rendering has come far and produced convincing results by considering the inner construction of the lens and how light interacts with it. However, obtaining a specific flare signature through these algorithms requires lens design expertise that the typical artist does not possess.

This thesis demonstrates how to obtain optical lens systems that achieve a desired flare effect, without requiring prior knowledge of lens construction. We achieve this through simplified controls and evolutionary algorithms. With this method, existing lenses can be tweaked, or even new ones can be built from scratch. The process abstracts the lens and algorithmic parameters, making these rendering algorithms accessible to a wider range of users. ... Read more

3D Gaussian splatting (3DGS) is an appealing implementation of novel view synthesis, with fast training and render times compared to related methods. However, per-frame sorting and front-to-back alpha compositing lead to a significant decline in performance for scenes with a high number of Gaussians. In particular, the alpha compositing step in the original implementation inefficiently handles overdraw. Therefore, dense clusters of Gaussians observed at a distance significantly increase frame times.


In this thesis, we propose a stochastic rendering approach that significantly improves performance in large scenes, by integrating principles from stochastic transparency and compute-based point cloud rendering. Each frame, we first compute a weight for each Gaussian based on its screen footprint and opacity. These weights are then used to construct an alias table on the GPU. We then splat points, each sampled from a Gaussian selected in proportion to its weight using the alias table, onto an interleaved depth-and-color buffer. By using an atomic operation, we ensure that the points closest to the camera are kept without sorting. To construct the final image, we average multiple samples per pixel, obtained across multiple point cloud passes. Although our approach introduces noise at low sample counts and minor artifacts due to the independent and identically distributed nature of the sampling, these effects can be mitigated using temporal anti-aliasing and rejection sampling respectively.

Our evaluations indicate that the render time of our method depends primarily on the density of Gaussians near the camera, rather than the total number in the scene. This allows our method to efficiently handle large-scale scenes. ... Read more

Augmented Reality (AR) offers significant potential to enhance surgical procedures by overlaying critical digital information directly onto the surgeon's view, improving spatial awareness and precision. Accurate real-time tracking of surgical instruments is fundamental to the success of AR navigation systems. However, implementing established tracking modalities, such as infrared (IR) tracking using retro-reflective markers, on newer Head-Mounted Displays (HMDs) like the Magic Leap 2 presents unique challenges due to its specific sensor configuration, notably the lack of accessible IR intensity data from its Time-of-Flight (ToF) sensor and the presence of IR-cut filters on its World Cameras.

This thesis addresses these challenges through the development, implementation, and evaluation of ISITLeap, a novel system designed specifically for IR instrument tracking on the Magic Leap 2 platform. The system employs a stereo vision approach, utilizing the Magic Leap 2's built-in World Cameras configured with short exposure times, synergistically combined with a custom-designed external 760 nm IR illumination module to detect standard passive retro-reflective markers. A comprehensive software pipeline was developed, incorporating image undistortion adapted for the Magic Leap 2's equidistant projection model, marker detection involving binarization, connected components analysis, and area filtering, 3D marker reconstruction using stereo triangulation with empirically derived coordinate system corrections to reconcile Magic Leap 2 SDK data with geometric requirements, and finally, 6-DoF tool pose estimation via point-based registration. Crucially, the entire software pipeline, from image acquisition to pose estimation, operates directly on the Magic Leap 2 headset, requiring no external computing hardware.

Experimental validation successfully demonstrated the feasibility of the ISITLeap system for real-time tracking. Static evaluations characterized the system's performance, identifying an optimal working range of approximately 25-50 cm. Dynamic tracking experiments, evaluated against an NDI optical tracker using an ArUco marker as a common reference frame, yielded an overall Mean Absolute Error (MAE) of approximately 7 mm. Crucially, error analysis revealed that the measured accuracy was significantly influenced by limitations inherent in the ArUco marker-based evaluation methodology, likely masking the intrinsic performance capabilities of the ISITLeap core tracking system.

This work presents a complete, functional hardware and software solution tailored to overcome the specific sensor constraints of the Magic Leap 2, enabling IR marker tracking on this platform. It provides a crucial performance benchmark under the defined evaluation conditions and demonstrates a viable technical pathway for developing precise AR surgical navigation applications for the Magic Leap 2 headset. While the achieved referenced accuracy highlights the need for improved evaluation techniques and further system refinement (e.g., via dynamic filtering and optimized hardware), ISITLeap establishes a foundational system for future advancements in Magic Leap 2-based AR surgical guidance. ... Read more

This thesis presents a computational framework aimed at enabling the analysis and modeling of boomerangs from example shapes. The goal is to provide a systematic and data-driven tool for boomerang design based on real-world geometries. A key challenge in this context is establishing accurate shape correspondences between handcrafted, asymmetric, and aerodynamically functional boomerangs.
To address this, the proposed pipeline integrates multiple components: landmark-based pre-alignment, boundary extraction using alpha shapes, curve parameterization, and Least Squares Conformal Mapping (LSCM) to compute surface correspondences. Building on these correspondences, the framework further incorporates principal component analysis (PCA) and Free-Form Deformation (FFD) to enable the generation of new shapes.
Experimental results show that the method achieves low Hausdorff and Chamfer distances and has been evaluated using area- and shear-based distortion metrics. Nonetheless, some localized inaccuracies - particularly near high-curvature regions - highlight areas for improvement in boundary handling and local control. Additional limitations include the reliance on manual landmark selection, the linearity of PCA, and the sensitivity of the pipeline to alpha shape parameters.
By combining elements of computational geometry with principles of aerodynamic design, this research bridges the gap between empirical craftsmanship and formal shape analysis. The resulting methodology not only enables the comparison and modeling of boomerangs but also lays the groundwork for future tools in boomerang shape exploration and design.

... Read more

Mirages are a visual phenomenon consisting of the appearance of a mirrored image of an object, without the presence of an actual mirror surface, due to light rays that are curved because of continuous refraction in the air, which relates to differences between the surface and ambient temperature.
As temperature is defined in space, while a typical standard rasterization pipeline only processes surfaces, these phenomena are difficult to reproduce.
Approximating the nonlinear ray path with ray marching becomes taxing due to the long light paths.
Current approaches use acceleration structures and have not been implemented in a rasterizer.

We present two methods that make dynamic real-time rendering of mirages possible, which fit well in the rasterization pipeline.
The first solution uses a second camera to capture surface temperature and normal information below the view ray, and approximates the nonlinear path of the ray in as few steps as possible.
The second method obtains the surface information in screen space instead, making it faster, but potentially less accurate in heterogeneous scenes.

Results show that both methods are capable of rendering mirages dynamically and in real-time when the surface is relatively flat.
Therefore, both methods, especially the second, faster method, could be used for the rendering of mirages on relatively flat faces, enabling real-time dynamic rendering of mirages in video games on those types of surfaces. ... Read more

This thesis presents the design and implementation of a GPU-based ray tracer capable of rendering NVIDIA's micro-meshes. The method operates entirely in two dimensions by projecting both rays and triangles onto a 2D domain. Within this domain, the projected ray traverses the micro-triangles, starting at subdivision level 0. The triangles are then recursively subdivided until the finest subdivision level is reached, at which point the micro-triangle intersected by the original 3D ray is identified. To accelerate traversal, the approach employs height field ray tracing, which restricts subdivision to regions where it is strictly necessary. In addition, the proposed method demonstrates a reduced memory footprint compared to traditional ray tracing, while maintaining acceptable runtime performance. Memory savings of up to approximately 50% are achieved relative to a fully tessellated representation. ... Read more

We propose a new method for simulating and rendering watery paint, which provides artists with a tool to produce watercolour-like textures on 3D models. Our particle-based system replicates most effects and techniques associated with watercolour, interactively and intuitively, as it is based on a physical fluid simulation. We show how our system can be applied to 3D surfaces by accurately compensating for distortions in the surface parametrisation. ... Read more

This dissertation investigates the use of deep-learning methods for user-guided content creation, manipulation, and exploration. It illustrates the potential of neural techniques to support working with large data collections and we illustrate our solutions through several applications. Regarding content generation, we propose an algorithm to produce material representations from a single image. We illustrate content manipulation with an approach to perform perceptually plausible interpolation and examine exploration in the context of interactive retrieval. For the latter, we show that features spaces are of high relevance to organize data and show the generality of this concept by proposing novel exploration methods for image and music collections. ... Read more

Despite the exponential growth in computational power, displaying highly detailed scenes remains challenging due to memory and performance limitations. One method of representing scenes is using voxels; the tiny cubic cells of a regular grid. In this dissertation, we are concerned with the Sparse Voxel Directed Acyclic Graph (SVDAG) data structure and how it can be used to improve large-scene rendering. The SVDAG is a hierarchical data structure that aims to reduce memory usage by identifying geometric redundancy. We address several challenges regarding this type of voxel compression. ... Read more

3D rendering is traditionally based on a tristimulus approximation, where all light, color, and spectral distributions are represented using three (RGB) values. For enhanced physical accuracy, spectral rendering algorithms can be employed. However, these methods are typically more computationally expensive and require scene and material data measured from the real world. With few exceptions, spectral rendering remains confined to academic research, with limited adoption in production pipelines due to the many challenges it poses. In this dissertation, we identify several of these challenges and propose practical solutions to each. ... Read more

Lag in Virtual Reality can be devastating to the enjoyment of participants. This paper reduces the lag in a physics system by implementing a subset of the architecture proposed by Loren Peitso and Don Brutzman. This subset contains an event based physics system together with a networking system. The collision system of the physics system has been heavily reduced in size, as to speedup development of the architecture.

This architecture seems promising and handles predictable cases with almost no noticeable lag. This architecture does not handle external events well, as the lag is noticeable. The `healing' process for unexpected external events should be researched further to make this architecture viable. ... Read more