Reproduce “Non-parametric Depth Distribution Modelling based Depth Inference for Multi-view Stereo” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiayu Yang, Jose M. Alvarez, Miaomiao Liu
Researchers developed a multi-view stereo method that models pixel depth using non-parametric distributions instead of unimodal assumptions, specifically addressing inaccuracies at object boundaries in coarse-to-fine pipelines. The approach reduced depth error on the most challenging DTU boundary regions (R0) to 6.48 mm, an improvement over PatchmatchNet's 9.54 mm.
Key insights
- The unimodal depth distribution assumption in cascaded MVS pipelines is a primary source of error, particularly at object boundaries and for small structures.
- Maintaining and branching multiple depth hypotheses from non-parametric distributions at coarse levels, rather than making early single-depth decisions, is critical for accurate refinement.
- Sparse cost aggregation, enabled by sparse 3D convolutions, is essential to process the irregular 3D data produced by pixel-wise depth branching, preserving geometric relationships.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.