Reliability-First 3D Reconstruction from Imperfect Consumer-Grade Handheld Captures
Most 3D reconstruction demos start after the input data is already clean. RobustCap3D studies what happens when the input capture is imperfect.
This repository will contain a reproducible 3D reconstruction reliability pipeline for consumer-grade handheld captures, comparing Raw, Masked and Quality-filtered + Masked inputs across SfM, neural rendering, geometric metrics and failure analysis.
Project initialization in progress.
- MVP: capture, quality gates, masking, COLMAP/PyCOLMAP, Nerfstudio, metrics and report.
- Technical extension: sparse-view sweeps, C++/OpenCV utilities, Mini-NeRF and advanced baselines.
- R&D extension: confidence-aware frame and mask selection.