LiDAR-EVS: Enhance Extrapolated View Synthesis for 3D Gaussian Splatting with Pseudo-LiDAR Supervision

ECCV 2026
Yiming Huang1,2,*, Xin Kang1,*, Sipeng Zhang1, Hongliang Ren2, Weihua Zhang1, Junjie Lai1,†
1NVIDIA    2The Chinese University of Hong Kong
* Equal contribution. † Corresponding author.
Contact: yimingh@nvidia.com, joeyk@nvidia.com, sipengz@nvidia.com, weihuaz@nvidia.com, julienl@nvidia.com
arXiv PDF Code

Extrapolated Right Shifted 3m NVS CD ↓48.6% vs. SplatAD

Extrapolated Left Shifted 3m NVS CD ↓47.2% vs. SplatAD

Interpolated NVS (Original Traj.) Comparable to origin

Each video visualizes bird's-eye-view LiDAR predictions from LiDARGS, NeurAD, SplatAD, and LiDAR-EVS. Our method preserves cleaner geometry and more complete structures under 3m left/right extrapolated trajectories, while maintaining performance comparable to the original trajectory for interpolated-view rendering.

Abstract

3D Gaussian Splatting (3DGS) has emerged as a powerful technique for real-time LiDAR and camera synthesis in autonomous driving simulation. However, simulating LiDAR with 3DGS remains challenging for extrapolated views beyond the training trajectory, as existing methods are typically trained on single-traversal sensor scans, suffer from severe overfitting and poor generalization to novel ego-vehicle paths. To enable reliable simulation of LiDAR along unseen driving trajectories without external multi-pass data, we present LiDAR-EVS, a lightweight framework for robust extrapolated-view LiDAR simulation in autonomous driving. Designed to be plug-and-play, LiDAR-EVS readily extends to diverse LiDAR sensors and neural rendering baselines with minimal modification. Our framework comprises two key components: (1) pseudo extrapolated-view point cloud supervision with multi-frame LiDAR fusion, view transformation, occlusion curling, and intensity adjustment; (2) spatially-constrained dropout regularization that promotes robustness to diverse trajectory variations encountered in real-world driving. Extensive experiments demonstrate that LiDAR-EVS achieves SOTA performance on extrapolated-view LiDAR synthesis across three datasets, making it a promising tool for data-driven simulation, closed-loop evaluation, and synthetic data generation in autonomous driving systems.

LiDAR-EVS teaser comparing extrapolated-view synthesis results.

LiDAR-EVS improves extrapolated-view LiDAR synthesis beyond the training trajectory.

Methodology

LiDAR-EVS pipeline.

Our framework consists of two key modules: Pseudo LiDAR Curation and Spatially Constrained Dropout. Pseudo LiDAR curation include the following steps: (1) Multi-frame fusion, (2) Extrapolated view transformation, (3) Occlusion curling, (4) Intensity adjustment. With the proposed framework, we can optimize the Gaussian scene representation to achieve robust LiDAR synthesis for both interpolated and extrapolated view rendering.

Experiment Results

Para-Lane Quantitative Results

Table 1: Para-Lane quantitative results.

Para-Lane Qualitative Results

Qualitative results on Para-Lane.

nuScenes Quantitative Results

Table 2: nuScenes quantitative results.

PandaSet Quantitative Results

Table 3: PandaSet quantitative results.

nuScenes Qualitative Results

Qualitative results on nuScenes.

Image Rendering Results

Table 4: Image rendering quantitative results.

RGB Rendering Qualitative Results

Qualitative RGB rendering results.

BibTeX

@misc{huang2026lidarevsenhanceextrapolatedview,
        title={LiDAR-EVS: Enhance Extrapolated View Synthesis for 3D Gaussian Splatting with Pseudo-LiDAR Supervision}, 
        author={Yiming Huang and Xin Kang and Sipeng Zhang and Hongliang Ren and Weihua Zhang and Junjie Lai},
        year={2026},
        eprint={2603.14763},
        archivePrefix={arXiv},
        primaryClass={cs.RO},
        url={https://arxiv.org/abs/2603.14763}, 
  }