Work

Structure, Not Just Magnitude, Drives the Damage from Misprojected LiDAR Depth Supervision in 3D Gaussian Splatting

Year: 2026

Type: software

Abstract: Analytic scenes with closed-form per-return co-visibility, used to separate the magnitude of cross-sensor depth-label error from its structure. Includes the 300-run analytic matrix, a 90-run nuScenes corroboration, a 90-run depth-sliced opacity probe of the damage mechanism with its per-view opacity maps, the 270 run records of the first pipeline version, and the scripts that apply the three frozen pre-analysis plans. The plan documents and the source-review audit records are not included.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.22930374

Authors: Zhao, Yifu, Zou, Xiaofan, Wei, Junhao, Li, Yanxiao, Li, Haochen, Im, Sio-Kei, Wang, Yapeng, Yang, Xu

Institutions: Macao Polytechnic University, Shanghai University

Language: English

Topic: 3D Shape Modeling and Analysis

Subfield: Computer Graphics and Computer-Aided Design

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7214166752

Cites:

Cited by:

Related to: 10