Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SARA: Scene-Aware Reconstruction Accelerator

Jee Won Lee, Hansol Lim, Minhyeok Im, Dohyeon Lee, Jongseong Brad Choi

TL;DR

SARA tackles the inefficiency of traditional SfM pair selection by replacing purely appearance-based candidate viewing with a geometry-driven scoring of reconstruction informativeness, defined as the product of overlap and parallax. It builds an Information-Weighted Spanning Tree (IWST) as a sparse, robust backbone and augments it with targeted edges for loop closure, long-baseline anchors, and weak-view reinforcement, dramatically reducing matcher calls from quadratic to quasi-linear. Across multiple modern detectors and NVS pipelines, SARA achieves up to ~50x speedups while maintaining or improving pose and reconstruction quality, with 3D Gaussian Splatting and SVRaster showing comparable downstream rendering performance to exhaustive matching. This approach enables scalable, geometry-aware SfM suitable for real-time or large-scale applications without sacrificing accuracy in downstream 3D reconstruction and view synthesis.

Abstract

We present SARA (Scene-Aware Reconstruction Accelerator), a geometry-driven pair selection module for Structure-from-Motion (SfM). Unlike conventional pipelines that select pairs based on visual similarity alone, SARA introduces geometry-first pair selection by scoring reconstruction informativeness - the product of overlap and parallax - before expensive matching. A lightweight pre-matching stage uses mutual nearest neighbors and RANSAC to estimate these cues, then constructs an Information-Weighted Spanning Tree (IWST) augmented with targeted edges for loop closure, long-baseline anchors, and weak-view reinforcement. Compared to exhaustive matching, SARA reduces rotation errors by 46.5+-5.5% and translation errors by 12.5+-6.5% across modern learned detectors, while achieving at most 50x speedup through 98% pair reduction (from 30,848 to 580 pairs). This reduces matching complexity from quadratic to quasi-linear, maintaining within +-3% of baseline reconstruction metrics for 3D Gaussian Splatting and SVRaster.

SARA: Scene-Aware Reconstruction Accelerator

TL;DR

SARA tackles the inefficiency of traditional SfM pair selection by replacing purely appearance-based candidate viewing with a geometry-driven scoring of reconstruction informativeness, defined as the product of overlap and parallax. It builds an Information-Weighted Spanning Tree (IWST) as a sparse, robust backbone and augments it with targeted edges for loop closure, long-baseline anchors, and weak-view reinforcement, dramatically reducing matcher calls from quadratic to quasi-linear. Across multiple modern detectors and NVS pipelines, SARA achieves up to ~50x speedups while maintaining or improving pose and reconstruction quality, with 3D Gaussian Splatting and SVRaster showing comparable downstream rendering performance to exhaustive matching. This approach enables scalable, geometry-aware SfM suitable for real-time or large-scale applications without sacrificing accuracy in downstream 3D reconstruction and view synthesis.

Abstract

We present SARA (Scene-Aware Reconstruction Accelerator), a geometry-driven pair selection module for Structure-from-Motion (SfM). Unlike conventional pipelines that select pairs based on visual similarity alone, SARA introduces geometry-first pair selection by scoring reconstruction informativeness - the product of overlap and parallax - before expensive matching. A lightweight pre-matching stage uses mutual nearest neighbors and RANSAC to estimate these cues, then constructs an Information-Weighted Spanning Tree (IWST) augmented with targeted edges for loop closure, long-baseline anchors, and weak-view reinforcement. Compared to exhaustive matching, SARA reduces rotation errors by 46.5+-5.5% and translation errors by 12.5+-6.5% across modern learned detectors, while achieving at most 50x speedup through 98% pair reduction (from 30,848 to 580 pairs). This reduces matching complexity from quadratic to quasi-linear, maintaining within +-3% of baseline reconstruction metrics for 3D Gaussian Splatting and SVRaster.
Paper Structure (27 sections, 14 equations, 5 figures, 4 tables)

This paper contains 27 sections, 14 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Contributions.
  3. Related Work
  4. Feature Extractors.
  5. Image Retrieval and Pair Selection.
  6. Feature Matching.
  7. Preliminaries
  8. Generalized Feature Detector.
  9. Generalized Feature Matcher.
  10. Geometric Informativeness.
  11. Methodology
  12. Overview
  13. Pre-matching Scorer
  14. Candidate Pruning.
  15. Information Weighted Spanning Tree (IWST)
  16. ...and 12 more sections

Figures (5)

  • Figure 1: System Overview of SARA: Scene-Aware Reconstruction Accelerator.
  • Figure 2: Pre-matching scorer.
  • Figure 3: Backbone augmentation after IWST.
  • Figure 4: Qualitative comparison of novel view synthesis results using 3D Gaussian Splatting across different SfM configurations.
  • Figure 5: Qualitative comparison of novel view synthesis results using SVRaster across different SfM configurations.