ViSE: A Systematic Approach to Vision-Only Street-View Extrapolation

TL;DR

ViSE introduces a four-stage pipeline combining data-driven initialization, geometric priors, generative modeling, and artifact removal to improve street-view extrapolation in autonomous driving, achieving top performance in a benchmark competition.

Contribution

The paper presents a novel systematic approach with a four-stage pipeline that significantly advances street-view extrapolation for autonomous driving.

Findings

Achieved first place in the RealADSim NVS track at ICCV 2025.

Outperformed existing methods with a final score of 0.441 on the benchmark.

Demonstrated robustness and accuracy in street-view extrapolation tasks.

Abstract

Realistic view extrapolation is critical for closed-loop simulation in autonomous driving, yet it remains a significant challenge for current Novel View Synthesis (NVS) methods, which often produce distorted and inconsistent images beyond the original trajectory. This report presents our winning solution which ctook first place in the RealADSim Workshop NVS track at ICCV 2025. To address the core challenges of street view extrapolation, we introduce a comprehensive four-stage pipeline. First, we employ a data-driven initialization strategy to generate a robust pseudo-LiDAR point cloud, avoiding local minima. Second, we inject strong geometric priors by modeling the road surface with a novel dimension-reduced SDF termed 2D-SDF. Third, we leverage a generative prior to create pseudo ground truth for extrapolated viewpoints, providing auxilary supervision. Finally, a data-driven adaptation network removes time-specific artifacts. On the RealADSim-NVS benchmark, our method achieves a final score of 0.441, ranking first among all participants.