Learning to fuse reconstruction and generative views for driving videos
PRG-Fusion: Orchestrating Generative Priors with Reconstruction Evidence for Driving View Synthesis
Computer Vision and Pattern Recognition
Summary
Creating realistic driving videos from new viewpoints is difficult because existing methods either produce inconsistent images or lose detail far from known camera paths. The authors propose a new system called PRG-Fusion that mixes different techniques depending on how well the current view can be reconstructed. It labels regions to keep the original look, fix structural errors, or fully generate new content. This approach improves image quality and consistency, especially when simulating driving scenes from perspectives the system hasn’t seen before.
view synthesisgenerative modelsneural renderingLiDARspatiotemporal consistency3D geometrytrajectoryvideo simulationclosed-loop simulation
Authors
Sipeng He, Jialei Chen, Zhen Fang, Dongchun Ren, Feng Zhao
Abstract
Synthesizing photorealistic driving videos along specified trajectories is essential for scalable closed-loop simulation. Reconstruction-based methods leverage neural rendering to synthesize geometrically consistent views, but often exhibit diverse artifacts and missing content when the viewpoint deviates from the training trajectory. In contrast, generative models can synthesize realistic views along arbitrary trajectories from vehicle sensor data, yet often struggle to maintain temporal and geometric consistency across frames. To combine the strengths of both, we propose PRG-Fusion, a framework for driving view synthesis that uses reconstruction evidence to orchestrate generative priors across regions. Specifically, we extract region-wise degradation evidence from reconstructed driving scenes and convert it into Preserve, Repair, and Generate (PRG) labels. At inference, these labels serve as a unified routing policy for region-aware spatiotemporal synthesis, orchestrating 3DGS appearance preservation, LiDAR-guided structural correction, and video-prior-driven content completion across Preserve, Repair, and Generate regions, respectively. We then follow a two-stage training paradigm, first establish geometric control from sparse LiDAR projections and subsequently learning appearance control from dense 3DGS renderings. Extensive experiments on Waymo demonstrate that PRG-Fusion achieves state-of-the-art overall performance in novel trajectory video synthesis, with superior visual quality and geometric fidelity while maintaining competitive view consistency under large trajectory shifts.