View-Graph Selection Framework for SfM

Abstract

View-graph is an essential input to large-scale structure from motion (SfM) pipelines. Accuracy and efficiency of large-scale SfM is crucially dependent on the input viewgraph. Inconsistent or inaccurate edges can lead to inferior or wrong reconstruction. Most SfM methods remove 'undesirable' images and pairs using several, fixed heuristic criteria, and propose tailor-made solutions to achieve specific reconstruction objectives such as efficiency, accuracy, or disambiguation. In contrast to these disparate solutions, we propose a single optimization framework that can be used to achieve these different reconstruction objectives with task-specific cost modeling. We also construct a very efficient network-flow based formulation for its approximate solution. The abstraction brought on by this selection mechanism separates the challenges specific to datasets and reconstruction objectives from the standard SfM pipeline and improves its generalization. This paper demonstrates the application of the proposed view-graph framework with standard SfM pipeline for two particular use-cases, (i) accurate and ghost-free reconstructions of highly ambiguous datasets using costs based on disambiguation priors, and (ii) accurate and efficient reconstruction of large-scale Internet datasets using costs based on commonly used priors.

View‐graph key‐subset extraction for efficient and robust structure from motion

“…However, the completeness and accuracy are still affected compared to using the full set. Shah et al (2018) proposed an optimisation‐based formulation to achieve different reconstruction objectives. The method improves efficiency with a comparable accuracy compared to the full view‐graph.…”

“…This redundancy slows methods that are sensitive to the number of images, such as COLMAP (Schonberger & Frahm, 2016). Second, redundancy with an uneven distribution usually occurs with insufficient baselines, which causes more unreliable 3D geometry in triangulation‐based SfM (Michelini & Mayer, 2020; Shah et al, 2018). Finally, an uneven distribution has a negative effect on bundle adjustment (BA) accuracy.…”

A Hybrid Global Image Orientation Method for Simultaneously Estimating Global Rotations and Global Translations

“…Based on these two criteria, the triplet compatibility indicator is formulated as max ( ∠ ( , 3×3 ) , ∠ ( , 180°) ). Finally, we employ a greedy triplet deleting scheme: starting with the triplet with the largest indicator, a triplet is deleted as long as the remaining triplets are still connected and no image is deleted from the photogrammetric block (see Appendix for more details on generating the OMCTS), note that we introduce our triplet selection process in a less sophisticated way and a more grounded graph theory based explanation is given by Shah et al (2018).…”

“…This can create problems, if rotations are incorrectly estimated. In addition, many researchers (Wilson and Snavely, 2014;Shah et al, 2018) also found that global methods are sensitive to outliers of relative orientations, since outliers are difficult to detect in global computations.…”

Spatial and Photometric Consistent Matching for Structure-from-Motion in Highly Ambiguous Scenes

“…However, large-scale scenes often contain ambiguous structures caused by repeating or symmetric elements and the reconstruction results in SfM can be collapsed by their false match pairs. Recently, some research [Shah et al 2018;Shen et al 2016] has attempted to use additional prior knowledge to solve this issue and improve the SfM performance. Nontheless, these methods are still too scene dependent and still cannot solve more challenging situations, such as stairwells.…”