Zero-shot depth map restoration from sparse infrastructure point clouds using diffusion models and prompted segmentation

doi:10.1016/j.autcon.2026.106839

Zero-shot depth map restoration from sparse infrastructure point clouds using diffusion models and prompted segmentation

2026-02-12

0

OA

AI

DOI：10.1016/j.autcon.2026.106839

Original

Save

View PDF

Abstract

En 中文

• Introduce a zero-shot framework, InfraDiffusion, to restore depth maps from masonry point clouds. • Propose a virtual camera projection for depth map generation from point clouds. • Adapt DDNM with boundary masking for depth image restorations using pre-trained diffusion models. • Improve segmentation metrics across five datasets using SAM segmentation.

Keywords:

InfraDiffusion

Diffusion models

Image restoration

Point clouds

Masonry structures

Depth maps

Semantic segmentation

Structural health monitoring

Zero-shot depth map restoration from sparse infrastructure point clouds using diffusion models and prompted segmentation

Abstract

AI Summary

Journal

Organization