EG 2024 - Posters

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/3606950

Browse

Now showing 1 - 11 of 11

VirtualVoxelCrowd: Rendering One Billion Characters at Real-Time
(The Eurographics Association, 2024) Yang, Jinyuan; Campbell, Abraham G.; Liu, Lingjie; Averkiou, Melinos
In this paper, we introduce VirtualVoxelCrowd, which aims to address the challenges of data scale and overdraw in massive crowd rendering applications. The approach leverages multiple levels of detail and multi-pass culling to reduce rendering workload and overdraw. VirtualVoxelCrowd supports rendering of up to one billion characters, achieving unprecedented scale on standard graphics hardware while rendering subpixel-level voxels to prevent the level of detail transition artifacts. This method offers significant improvements in handling massive animated crowd visualization, establishing a new possibility for dynamic, large-scale scene rendering.
Topological Data Structure for Computer Graphics
(The Eurographics Association, 2024) Fábián, Gábor; Liu, Lingjie; Averkiou, Melinos
This research is motivated by the following well-known contradiction. In computer-aided design or modeling tasks, we generally represent surfaces using edge-based data structures as winged edge [Bau75], half-edge [MP78] [CP98], or quad-edge [GS85]. In contrast, real-time computer graphics represents surfaces with face-vertex meshes, since for surface rendering, there is no need for the explicit representation of edges. In this research we introduce a novel data structure for representation of triangle meshes. Our representation is based on the concept of face-vertex meshes with adjacencies, but we use some extra information and new ideas that greatly simplify the implementation of algorithms.
ShapeVerse: Physics-based Characters with Varied Body Shapes
(The Eurographics Association, 2024) Vyas, Bharat; O'Sullivan, Carol; Liu, Lingjie; Averkiou, Melinos
Computer animation of realistic human characters remains a significant challenge. This work used deep reinforcement learning to generate physics-based characters with diverse body shapes. We aimed to replicate reference motions like walking or jogging while considering individual variations in body shape and mass. Reference motions served as training targets, accounting for differences in shape parameters to accommodate mass variations. This method produced animations that accurately capture human motion details, leading to diverse and lifelike character performances.
Reconstruction of Sparse Hyperspectral BRDF Measurements Preserving Their Physical and Topological Properties
(The Eurographics Association, 2024) Margall, François; Ceolato, Romain; Coiro, Eric; Mavromatis, Sébastien; Pacanowski, Romain; Liu, Lingjie; Averkiou, Melinos
The measurement of hyperspectral bidirectional reflectance distribution function (BRDF) of a material is a key issue in physically based spectral rendering. We present here a device for measuring BRDF, enabling high spectral sampling (sub-nanometric from 450 nm to 1100 nm) with the counterpart of sparser angular sampling. To overcome this problem, we propose an interpolation method that respects the physical and topological properties of the BRDF by construction. The characteristic properties of the material deduced from the interpolated data set correspond to the reference values obtained using dedicated measuring instruments.
Distributed Surface Reconstruction
(The Eurographics Association, 2024) Marin, Diana; Komon, Patrick; Ohrhallinger, Stefan; Wimmer, Michael; Liu, Lingjie; Averkiou, Melinos
Recent advancements in scanning technologies and their rise in availability have shifted the focus from reconstructing surfaces from point clouds of small areas to large, e.g., city-wide scenes, containing massive amounts of data. We adapt a surface reconstruction method to work in a distributed fashion on a high-performance cluster, reconstructing datasets with millions of vertices in seconds. We exploit the locality of the connectivity required by the reconstruction algorithm to efficiently divide-andconquer the problem of creating triangulations from very large unstructured point clouds.
Behavioral Landmarks: Inferring Interactions from Data
(The Eurographics Association, 2024) Lemonari, Marilena; Charalambous, Panayiotis; Panayiotou, Andreas; Chrysanthou, Yiorgos; Pettré, Julien; Liu, Lingjie; Averkiou, Melinos
We aim to unravel complex agent-environment interactions from trajectories, by explaining agent paths as combinations of predefined basic behaviors. We detect trajectory points signifying environment-driven behavior changes, ultimately disentangling interactions in space and time; our framework can be used for environment synthesis and authoring, shown by our case studies.
Dense 3D Gaussian Splatting Initialization for Sparse Image Data
(The Eurographics Association, 2024) Seibt, Simon; Chang, Thomas Vincent Siu-Lung; von Rymon Lipinski, Bartosz ; Latoschik, Marc Erich; Liu, Lingjie; Averkiou, Melinos
This paper presents advancements in novel-view synthesis with 3D Gaussian Splatting (3DGS) using a dense and accurate SfM point cloud initialization approach. We address the challenge of achieving photorealistic renderings from sparse image data, where basic 3DGS training may result in suboptimal convergence, thus leading to visual artifacts. The proposed method enhances precision and density of initially reconstructed point clouds by refining 3D positions and extrapolating additional points, even for difficult image regions, e.g. with repeating patterns and suboptimal visual coverage. Our contributions focus on improving ''Dense Feature Matching for Structure-from-Motion'' (DFM4SfM) based on a homographic decomposition of the image space to support 3DGS training: First, a grid-based feature detection method is introduced for DFM4SfM to ensure a welldistributed 3D Gaussian initialization uniformly over all depth planes. Second, the SfM feature matching is complemented by a geometric plausibility check, priming the homography estimation and thereby improving the initial placement of 3D Gaussians. Experimental results on the NeRF-LLFF dataset demonstrate that this approach achieves superior qualitative and quantitative results, even for fewer views, and the potential for a significantly accelerated 3DGS training with faster convergence.
Comparing NVIDIA RTX and a Novel Voxel-Space Ray Marching Approach as Global Illumination Solutions
(The Eurographics Association, 2024) Erlich, Oren; Aristizabal, Sarah; Li, Lucas; Woodard, Brandon; Humer, Irene; Eckhardt, Christian; Liu, Lingjie; Averkiou, Melinos
In this work, we investigate the performance-as well as the quality difference-between the state of the art NVIDIA DXR ray tracing pipeline and a voxelspace ray marching (VSRM). In order to maintain an acceptable quality image outcome, as well as frame-rate, for tested low numbers of rays from one to 32, we use a simple denoiser. We show a similar quality outcome and less progressive dependency on the number of rays for VSRM compared with DXR.
EUROGRAPHICS 2024: Posters Frontmatter
(Eurographics Association, 2024) Liu, Lingjie; Averkiou, Melinos; Liu, Lingjie; Averkiou, Melinos
From Few to Full: High-Resolution 3D Object Reconstruction from Sparse Views and Unknown Poses
(The Eurographics Association, 2024) Yao, Grekou; Mavromatis, Sebastien; Mari, Jean-Luc; Liu, Lingjie; Averkiou, Melinos
Recent progress in 3D reconstruction has been driven by generative models, moving from traditional multi-view dependence to single-image diffusion model based techniques. However, these innovative approaches often face challenges with sparse view scenarios, requiring known poses or template shapes, often failing in high-resolution reconstructions. Addressing these issues, we introduce the ''F2F'' (Few to Full) framework, designed for crafting high-resolution 3D models from few views and unknown camera poses, creating fully realistic 3D objects without external constraints. F2F employs a hybrid approach, optimizing both implicit and explicit representations through a unique pipeline involving a pretrained diffusion model for pose estimation, a deformable tetrahedra grid for feature volume construction, and an MLP (neural network) for surface optimization. Our method sets a new standard by ensuring surface geometry, topology, and semantic consistency through differentiable rendering, aiming for a comprehensive solution in 3D reconstruction from sparse views.
Interactive VPL-based Global Illumination on the GPU Using Adaptive Fuzzy Clustering
(The Eurographics Association, 2024) Colom, Arnau; Marques, Ricardo; Santos, Luís Paulo; Liu, Lingjie; Averkiou, Melinos
Physically-based synthesis of high quality imagery results in a significant workload, which makes interactive rendering a very challenging task. Our approach to achieve such interactive frame rates while accurately simulating global illumination phenomena entails developing a Virtual Point Lights (VPL) ray tracer that runs entirely in the GPU. Our performance guarantees arise from clustering both shading points and VPLs and computing visibility only among clusters' representatives. Previous approaches to the same problem resort to K-means clustering, which requires the user to specify the number of clusters; a rather unintuitive requirement. We propose an innovative massively parallel, GPU-efficient, Quality-Threshold clustering algorithm, which requires the user to specify a quality parameter. The algorithm dynamically adjusts the number of clusters depending both on the specified quality threshold and on camera-geometry conditions during execution.

Browse

Browsing EG 2024 - Posters by Issue Date

Results Per Page

Sort Options