Browse Teach Curate

Sparse-View Indoor Neural Surface Reconstruction via Inter-Image Matching Priors

A neural implicit surface reconstruction approach that recovers indoor scene geometry from only a handful of views. It replaces scale-ambiguous monocular depth priors with a depth prior derived from inter-image feature matching, enforces cross-view consistency through a reprojection loss over a signed distance function, and filters matching errors using an angular-separation filter and an epipolar (Sampson-distance) weighting function.

Data Text

Sparis: Neural Implicit Surface Reconstruction of Indoor Scenes from Sparse Views Yulun Wu1, 2*

Sparis is a neural implicit surface reconstruction method for recovering indoor scene geometry from only a sparse set of input views. Prior neural-implicit methods use monocular depth/normal priors b…