SimpleProc: Fully Procedural Synthetic Data from Simple Rules for Multi-View Stereo

TL;DR

SimpleProc is a procedural data generator for multi-view stereo that uses simple rules and NURBS to produce high-quality training data, outperforming manual datasets in several benchmarks.

Contribution

We introduce SimpleProc, a fully procedural multi-view stereo dataset generator that achieves superior performance with fewer rules and less manual curation.

Findings

At 8,000 images, outperforms manually curated datasets.

Scaling to 352,000 images yields comparable or better results than 692,000 images.

Source code and data are publicly available.

Abstract

In this paper, we explore the design space of procedural rules for multi-view stereo (MVS). We demonstrate that we can generate effective training data using SimpleProc: a new, fully procedural generator driven by a very small set of rules using Non-Uniform Rational Basis Splines (NURBS), as well as basic displacement and texture patterns. At a modest scale of 8,000 images, our approach achieves superior results compared to manually curated images (at the same scale) sourced from games and real-world objects. When scaled to 352,000 images, our method yields performance comparable to--and in several benchmarks, exceeding--models trained on over 692,000 manually curated images. The source code and the data are available at https://github.com/princeton-vl/SimpleProc.