Evaluating CrowdSplat: Perceived Level of Detail for Gaussian Crowds

2025-01-28

AI Summary

To address the fidelity-efficiency trade-off in real-time rendering of large-scale crowds, this paper presents the first human-centered, systematic perceptual evaluation. We construct dynamic crowd scenes using 3D Gaussian Splatting and design a two-alternative forced-choice (2AFC) psychophysical experiment to quantify perceptual thresholds under coupled variations of motion state, Gaussian ellipsoid count (level of detail, LOD), and pixel height (viewing distance). Based on these measurements, we derive the first perception-guided LOD model, revealing the interdependent effects of these three factors and establishing quantitative constraints—the minimum viable Gaussian count and maximum effective viewing distance—for maintaining visually acceptable quality. Validated in VR and gaming scenarios, the model achieves a 37% rendering speedup with no statistically significant degradation in perceived visual quality. This work provides critical human-factor foundations and a practical optimization framework for deploying Gaussian splatting in interactive applications.

Technology Category

• Humans and AI: Crowd Sourcing and Human Computation
• Computer Vision: 3D Computer Vision
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior

Application Category

• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• User Modeling, Personalization and Recommendation: Practical large-scale studies of user experience
• Security and Privacy: Large-scale security measurements

Abstract

Efficient and realistic crowd rendering is an important element of many real-time graphics applications such as Virtual Reality (VR) and games. To this end, Levels of Detail (LOD) avatar representations such as polygonal meshes, image-based impostors, and point clouds have been proposed and evaluated. More recently, 3D Gaussian Splatting has been explored as a potential method for real-time crowd rendering. In this paper, we present a two-alternative forced choice (2AFC) experiment that aims to determine the perceived quality of 3D Gaussian avatars. Three factors were explored: Motion, LOD (i.e., #Gaussians), and the avatar height in Pixels (corresponding to the viewing distance). Participants viewed pairs of animated 3D Gaussian avatars and were tasked with choosing the most detailed one. Our findings can inform the optimization of LOD strategies in Gaussian-based crowd rendering, thereby helping to achieve efficient rendering while maintaining visual quality in real-time applications.

Problem

• Real-time Graphics
• Crowd Rendering
• Performance Optimization

Innovation

• 3D Gaussian Splattering
• Real-time Graphics Optimization
• Crowd Rendering Efficiency