Recent advances in 3D Gaussian Splatting (3DGS) have focused on accelerating optimization while preserving reconstruction quality. However, many proposed methods entangle implementation-level improvements with fundamental algorithmic modifications or trade performance for fidelity, leading to a fragmented research landscape that complicates fair comparison. In this work, we consolidate and evaluate the most effective and broadly applicable strategies from prior 3DGS research and augment them with several novel optimizations. We further investigate underexplored aspects of the framework, including numerical stability, Gaussian truncation, and gradient approximation. The resulting system, Faster-GS, provides a rigorously optimized algorithm that we evaluate across a comprehensive suite of benchmarks. Our experiments demonstrate that Faster-GS achieves up to 5$\times$ faster training while maintaining visual quality, establishing a new cost-effective and resource efficient baseline for 3DGS optimization. Furthermore, we demonstrate that optimizations can be applied to 4D Gaussian reconstruction, leading to efficient non-rigid scene optimization.

Explicit primitive-based radiance fields such as 3D Gaussian Splatting typically model view-dependent appearance using low-order spherical harmonics (SH). Although efficient to evaluate, SH coefficients dominate per-primitive storage and memory traffic, while their band-limited basis restricts angular detail. We present a thorough, end-to-end comparison of SH and recent spherical appearance models and introduce an implicit alternative that decodes compact per-primitive latent codes using a tiny shared MLP. We integrate all models into the same optimized pipeline, fusing their forward and backward passes into a differentiable CUDA rasterizer and provide a portable WebGL viewer for laptop and mobile GPUs. Our evaluation across reconstruction quality, memory use, and optimization and rendering performance shows that recent spherical models offer the strongest overall quality-efficiency trade-off. Our neural representation is the most compact model evaluated and, compared to third-degree SH, reduces the per-primitive appearance footprint from 192 to 28 bytes, accelerates optimization by 1.3×, while improving reconstruction quality. We further analyze how appearance parametrization shapes optimization, identifying differences in recovered geometry and the tendency of expressive models to absorb non-static scene content. Together, our framework and analysis provide practical guidance for replacing SH beyond what image metrics alone can capture.

Ray-axis aligned bounding box intersection tests play a crucial role in the runtime performance of many rendering applications, driven not by complexity but mainly by the volume of tests required. While existing solutions were believed to be pretty much optimal in terms of runtime on current hardware, our paper introduces a new intersection test requiring fewer arithmetic operations compared to all previous methods. By transforming the ray we eliminate the need for one third of the traditional bounding-slab tests and achieve a speed enhancement of approximately 13.8% or 10.9%, depending on the compiler. We present detailed runtime analyses in various scenarios.

Several studies have examined the potential of immersive technologies, such as Virtual Reality (VR), to enhance the effectiveness of memorization. However, existing research has not specifically focused on individuals with Attention Deficit Hyperactivity Disorder (ADHD), and only a limited number of studies have examined the effectiveness of Memory Palace (MP) as a memorization aid for this population, who often experience working memory impairments. To address this gap, we recruited participants with an official diagnosis of ADHD and conducted an experiment in which we investigated the impact of both the Traditional MP technique and a VR version to assess their impact on a memorization task. Our findings indicate that the effectiveness of a VR-based MP might be influenced by prior experience with VR systems, the level of familiarity with the virtual environment, and the MP technique. Nevertheless, our results demonstrate that the MP technique has the potential to enhance recall performance in individuals diagnosed with ADHD.

Video calls have become an essential part of remote work. They enable employees to collaborate from different locations, including their homes. Transmitting video from the personal living environment, however, poses a privacy risk: Colleagues may gain insight into private information through details in the background. To limit this risk, video conferencing services implement virtual backgrounds that conceal the real environment during a video call. Unfortunately, this protection suffers from imperfections and pixels from the environment occasionally become visible. In this paper, we investigate this privacy leak. We analyze the virtual background techniques used in two major video conferencing services (Zoom and Google) and determine how pixels of the environment leak. Based on this analysis, we propose a reconstruction attack: This attack removes the virtual background by re-purposing the video conferencing software and uses semantic segmentation to filter out the video caller. As a result, only pixels leaking from the environment remain and can be aggregated into a reconstructed image. We examine the efficacy of this attack in a quantitative and qualitative evaluation. In comparison to previous studies, our attack recovers at least 53% more leaked pixels from a video call, exposing larger areas of the environment. We thus conclude that virtual backgrounds currently do not provide an adequate protection in practice.

We introduce a new approach for reconstruction and novel view synthesis of unbounded real-world scenes. In contrast to previous methods using either volumetric fields, grid-based models, or discrete point cloud proxies, we propose a hybrid scene representation, which implicitly encodes the geometry in a continuous octree-based probability field and view-dependent appearance in a multi-resolution hash grid. This allows for extraction of arbitrary explicit point clouds, which can be rendered using rasterization. In doing so, we combine the benefits of both worlds and retain favorable behavior during optimization: Our novel implicit point cloud representation and differentiable bilinear rasterizer enable fast rendering while preserving the fine geometric detail captured by volumetric neural fields. Furthermore, this representation does not depend on priors like structure-from-motion point clouds. Our method achieves state-of-the-art image quality on common benchmarks. Furthermore, we achieve fast inference at interactive frame rates, and can convert our trained model into a large, explicit point cloud to further enhance performance.

In this article, we introduce a playful learning method for teaching computer graphics, grounded in the three traditional learning theories of flow, constructivism, and cognitive load. Our approach consists of two main stages: first, the theoretical concepts are delivered through traditional lectures, and then, students engage in practical exercises designed to reduce cognitive load, promote flow, and provide hands-on experience with complex topics. We tested this method in a second-year undergraduate computer graphics course focused on transformations. Feedback from students and the quality of their final projects indicate a positive reception to this playful learning approach. To incorporate this approach into other computer graphics courses and more easily extend it to other topics, we also give an overview of freely available online tools anyone can use.

3D Gaussian Splats (3DGS) have proven a versatile rendering primitive, both for inverse rendering as well as real-time exploration of scenes. In these applications, coherence across camera frames and multiple views is crucial, be it for robust convergence of a scene reconstruction or for artifact-free fly-throughs. Recent work started mitigating artifacts that break multi-view coherence, including popping artifacts due to inconsistent transparency sorting and perspective-correct outlines of (2D) splats. At the same time, real-time requirements forced such implementations to accept compromises in how transparency of large assemblies of 3D Gaussians is resolved, in turn breaking coherence in other ways. In our work, we aim at achieving maximum coherence, by rendering fully perspective-correct 3D Gaussians while using a high-quality approximation of accurate blending, hybrid transparency, on a per-pixel level, in order to retain real-time frame rates. Our fast and perspectively accurate approach for evaluation of 3D Gaussians does not require matrix inversions, thereby ensuring numerical stability and eliminating the need for special handling of degenerate splats, and the hybrid transparency formulation for blending maintains similar quality as fully resolved per-pixel transparencies at a fraction of the rendering costs. We further show that each of these two components can be independently integrated into Gaussian splatting systems. In combination, they achieve up to 2x higher frame rates, 2x faster optimization, and equal or better image quality with fewer rendering artifacts compared to traditional 3DGS on common benchmarks.

Implicit Neural Point Cloud (INPC) is a recent hybrid representation that combines the expressiveness of neural fields with the efficiency of point-based rendering, achieving state-of-the-art image quality in novel view synthesis. However, as with other high-quality approaches that query neural networks during rendering, the practical usability of INPC is limited by comparatively slow rendering. In this work, we present a collection of optimizations that significantly improve both the training and inference performance of INPC without sacrificing visual fidelity. The most significant modifications are an improved rasterizer implementation, more effective sampling techniques, and the incorporation of pre-training for the convolutional neural network used for hole-filling. Furthermore, we demonstrate that points can be modeled as small Gaussians during inference to further improve quality in extrapolated, e.g., close-up views of the scene. We design our implementations to be broadly applicable beyond INPC and systematically evaluate each modification in a series of experiments. Our optimized INPC pipeline achieves up to 25

Dynamic reconstruction and spatiotemporal novel-view synthesis of non-rigidly deforming scenes recently gained increased attention. While existing work achieves impressive quality and performance on multi-view or teleporting camera setups, most methods fail to efficiently and faithfully recover motion and appearance from casual monocular captures. This paper contributes to the field by introducing a new method for dynamic novel view synthesis from monocular video, such as casual smartphone captures. Our approach represents the scene as a dynamic neural point cloud, an implicit time-conditioned point distribution that encodes local geometry and appearance in separate hash-encoded neural feature grids for static and dynamic regions. By sampling a discrete point cloud from our model, we can efficiently render high-quality novel views using a fast differentiable rasterizer and neural rendering network. Similar to recent work, we leverage advances in neural scene analysis by incorporating data-driven priors like monocular depth estimation and object segmentation to resolve motion and depth ambiguities originating from the monocular captures. In addition to guiding the optimization process, we show that these priors can be exploited to explicitly initialize our scene representation to drastically improve optimization speed and final image quality. As evidenced by our experimental evaluation, our dynamic point cloud model not only enables fast optimization and real-time frame rates for interactive applications, but also achieves competitive image quality on monocular benchmark sequences. Our code and data are available online: https://moritzkappel.github.io/projects/dnpc/.

In this paper we propose SPaGS, a high-quality, real-time free-viewpoint rendering approach from 360-degree panoramic images. While existing methods building on Neural Radiance Fields or 3D Gaussian Splatting have difficulties to achieve real-time frame rates and high-quality results at the same time, SPaGS combines the advantages of an explicit 3D Gaussian-based scene representation and ray casting-based rendering to attain fast and accurate results. Central to our new approach is the exact calculation of axis-aligned bounding boxes for spherical images that significantly accelerates omnidirectional ray casting of 3D Gaussians. We also present a new dataset consisting of ten real-world scenes recorded with a drone that incorporates both calibrated 360-degree panoramic images as well as perspective images captured simultaneously, i.e., with the same flight trajectory. Our evaluation on this new dataset as well as established benchmarks demonstrates that SPaGS excels over state-of-the-art methods in terms of both rendering quality and speed.

With the advent of holographic near-eye displays, the need for rendering algorithms that output holograms instead of color images emerged. These holograms usually encode phase maps that alter the phase of coherent light sources such that images result from diffraction effects. While common approaches rely on translating the output of traditional rendering systems to holograms in a post processing step, we instead developed a rendering system that can directly output a phase map to a Spatial Light Modulator (SLM). Our hardware-ray-traced sparse point distribution, and depth mapping enable rapid hologram generation, allowing for high-quality time-multiplexed holography for real-time content. Additionally, our system is compatible with foveated rendering which enables further performance optimizations.

A major factor resulting in cybersickness is the feeling of self-motion experienced when viewing a moving scene in Virtual Reality (VR). Current research indicates that this effect is largely created by motion in the periphery. To discover why this is the case, we investigate the influence of temporal frequency and eccentricity of a stimulus on the magnitude of perceived velocity in the periphery. Based on the perception of two-dimensional stimuli on a wide field-of-view display, we build a model to predict the scaling factor by which the perceived velocity of visual patterns deviates from the physical velocity. Further, our exploratory findings indicate no impact of gaze type on the results, suggesting our model works for both fixation and smooth pursuit scenarios. In an additional pilot study in VR, we test the accuracy of the model to predict unnoticeable object motion adaptation in 3D virtual worlds and find positive indications for a similar effect.

Virtual reality has ushered in a revolutionary era of immersive content perception. However, a persistent challenge in dynamic environments is the occurrence of cybersickness arising from a conflict between visual and vestibular cues. Prior techniques have demonstrated that limiting illusory self-motion, so-called vection, by blurring the peripheral part of images, introducing tunnel vision, or altering the camera path can effectively reduce the problem. Unfortunately, these methods often alter the user's experience with visible changes to the content. In this paper, we propose a new technique for reducing vection and combating cybersickness by subtly lowering the screen-space speed of objects in the user's peripheral vision. The method is motivated by our hypothesis that small modifications to the objects' velocity in the periphery and geometrical distortions in the peripheral vision can remain unnoticeable yet lead to reduced vection. This paper describes the experiments supporting this hypothesis and derives its limits. Furthermore, we present a method that exploits these findings by introducing subtle, screen-space geometrical distortions to animation frames to counteract the motion contributing to vection. We implement the method as a real-time post-processing step that can be integrated into existing rendering frameworks. The final validation of the technique and comparison to an alternative approach confirms its effectiveness in reducing cybersickness.

This paper presents a novel method for accelerating the computationally intensive process of point-based holography using consumer grade hardware. By leveraging the parallel processing capabilities of graphics processing units (GPUs) and implementing optimization techniques, the proposed method significantly reduces the time required to generate complex holograms. A comprehensive analysis - including benchmarks and comparative studies - demonstrates the efficiency and effectiveness of this approach. Our findings offer promising implications for real-time applications in virtual reality, and other fields that require rapid and accurate holographic rendering. (c) 2024 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement

Surrounding lighting conditions cannot always be sufficiently controlled during videoconferences, yielding situations in which disturbing reflections might appear on the participants glasses. In this article, we present a retrained neural network to convincingly reduce such reflections. For real time performance we propose an asynchronous processing pipeline accompanied by a head pose-based caching strategy to reuse intermediate processing results. The implementation as virtual webcam allows the system to be used with arbitrary videoconferencing systems.

This paper presents a point-based, neural rendering approach for complex real-world objects from a set of photographs. Our method is specifically geared towards representing fine detail and reflective surface characteristics at improved quality over current state-of-the-art methods. From the photographs, we create a 3D point model based on optimized neural feature points located on a regular grid. For rendering, we employ view-dependent spherical harmonics shading, differentiable rasterization, and a deep neural rendering network. By combining a point-based approach and novel regularizers, our method is able to accurately represent local detail such as fine geometry and high-frequency texture while at the same time convincingly interpolating unseen viewpoints during inference. Our method achieves about 7 frames per second at 800 × 800 pixel output resolution on commodity hardware, putting it within reach for real-time rendering applications.

Tracking errors severely impact the effectiveness of augmented reality display techniques for indoor navigation. In this work we take a look at the sources of error and accuracy of existing tracking technologies. We derive important design criteria for robust display techniques and present objective criteria. These serve evaluation of indoor navigation techniques without or in preparation of quantitative user studies. Based on these criteria we propose a new error tolerant display technique called Bending Words, where words move along the navigation path guiding the user. Bending Words outranks the other evaluated display techniques in many of the tested criteria and provides a robust, error-tolerant alternative to established augmented reality indoor navigation display techniques.