Local-Global Joint Modeling for Point Cloud Super-Resolution Guided by Multi-Scale Spatial Priors

概要

Range-image-based point cloud super-resolution provides an effective way to reconstruct sparse outdoor point clouds by projecting low-resolution LiDAR scans into a regular 2D representation. However, current methods still face three challenges. First, range images exhibit pronounced vertical nonuniformity, with sparse observations in distant regions and denser observations in nearby regions. Some cross-scale fusion mechanisms do not account for the varying reliability of features at different vertical positions, allowing invalid values and noise from distant sparse regions to propagate into the decoder. Second, a single convolutional or sequence-modeling architecture often struggles to preserve both local high-frequency geometric details and global long-range dependencies, resulting in over-smoothed reconstructions. Third, smooth transition depths near foreground-background boundaries can produce spike artifacts when the 2D range image is back-projected into 3D. To address these issues, we propose a point cloud super-resolution method based on multi-scale spatial priors and joint local-global modeling. The design comprises a spatially gated fusion module, a multi-granularity local-global cooperative perception module, and 2D median-residual denoising. Experiments on KITTI-360 and nuScenes demonstrate effective reconstruction, with IoU scores of 0.4703 and 0.3593, respectively, supporting the feasibility of the proposed method for range-image super-resolution.

参考リンク

• Paper ↗

リンク先のライセンス・再現性・データ条件をご確認ください。自動分類や実装の目安は保証ではありません。

実装を始めるプロンプト

参考リンクを確認したうえで、次の内容をコピーして利用できます。

# 論文・記事
Local-Global Joint Modeling for Point Cloud Super-Resolution Guided by Multi-Scale Spatial Priors

# 概要
Range-image-based point cloud super-resolution provides an effective way to reconstruct sparse outdoor point clouds by projecting low-resolution LiDAR scans into a regular 2D representation. However,…

# 原典
https://arxiv.org/abs/2611.02786

上記の手法を実装してください。実行前に依存関係とライセンスを確認してください。