EVolSplat: Efficient Volume-based Gaussian Splatting for Urban View Synthesis

2025-03-26

AI Summary

Existing NeRF and 3D Gaussian Splatting (3DGS) methods for urban novel-view synthesis suffer from scene-specific optimization, slow inference, multi-view inconsistency, and content duplication. To address these limitations, this paper proposes Volumetric Gaussian Splatting (VGS), a feed-forward volumetric framework. Its core contributions are: (1) a unified 3D convolutional network that directly predicts spatio-temporal Gaussian parameters across frames; (2) a joint optimization strategy integrating noise-aware depth-guided initialization, 3D geometric refinement, and 2D texture-driven shading; and (3) a deformable hemispherical sky background model. Evaluated on KITTI-360 and Waymo, VGS achieves state-of-the-art rendering quality among feed-forward 3DGS/NeRF approaches, enables real-time rendering (≥30 FPS), and significantly improves cross-scene generalization and geometry-appearance consistency.

Technology Category

• Computer Vision: 3D Computer Vision
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Virtualization and resource management in Web systems and infrastructures
• Search and Retrieval-Augmented AI: Vertical and domain-specific search

Abstract

Novel view synthesis of urban scenes is essential for autonomous driving-related applications.Existing NeRF and 3DGS-based methods show promising results in achieving photorealistic renderings but require slow, per-scene optimization. We introduce EVolSplat, an efficient 3D Gaussian Splatting model for urban scenes that works in a feed-forward manner. Unlike existing feed-forward, pixel-aligned 3DGS methods, which often suffer from issues like multi-view inconsistencies and duplicated content, our approach predicts 3D Gaussians across multiple frames within a unified volume using a 3D convolutional network. This is achieved by initializing 3D Gaussians with noisy depth predictions, and then refining their geometric properties in 3D space and predicting color based on 2D textures. Our model also handles distant views and the sky with a flexible hemisphere background model. This enables us to perform fast, feed-forward reconstruction while achieving real-time rendering. Experimental evaluations on the KITTI-360 and Waymo datasets show that our method achieves state-of-the-art quality compared to existing feed-forward 3DGS- and NeRF-based methods.

Problem

• Efficient urban scene synthesis for autonomous driving
• Overcoming slow per-scene optimization in NeRF and 3DGS
• Resolving multi-view inconsistencies in feed-forward 3DGS methods

Innovation

• Uses 3D convolutional network for Gaussian prediction
• Initializes Gaussians with noisy depth predictions
• Employs hemisphere model for distant views