Divide and Conquer: Efficient Large-Scale Structure from Motion Using Graph Partitioning

Bhowmick, Brojeshwar; Patra, Suvam; Chatterjee, A.; Govindu, Venu Madhav; Banerjee, Subhashis

doi:10.1007/978-3-319-16808-1_19

Cited by 24 publications

(38 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In contrast, global SfM techniques [8,10,25] are able to compute all camera poses simultaneously, leading to extremely efficient solvers for large-scale problems; however, these methods lack robustness and are typically less accurate than incremental SfM. Alternatively, hierarchical SfM methods compute 3D reconstructions by first breaking up the input images into clusters that are individually reconstructed then merged together into a common 3D coordinate system [2]. Typically, bundle adjustment is run each time clusters are merged.…”

Section: Related Workmentioning

confidence: 99%

“…We generated uniformly random camera configurations that placed cameras (i.e., ray origins) in the cube [−1, 1] × [−1, 1] × [−1, 1] around the origin. The 3D points were randomly placed in the volume [−1, 1] × [−1, 1] × [2,4]. Ray directions were computed as unit vectors from camera origins to 3D points.…”

Section: Numerical Stabilitymentioning

confidence: 99%

“…We begin by partitioning the input dataset into subsets using normalized cuts [18] similar to Bhowmick et al [2]. The size of the subsets depends on the size of the particular dataset, but typically the subsets contain between 50 and 250 cameras.…”

Section: Incremental Sfm With Distributed Camerasmentioning

confidence: 99%

“…We compare our SfM pipeline to Incremental SfM (Vi-sualSfM [26]), the DISCO pipeline of Crandall et al [6], and the hierarchical SfM pipeline of Bhowmick et al [2] run on several large-scale SfM datasets and show the results in Table 2. All methods were run on a desktop machine with a 3.4GHz i7 processor and 16GB of RAM.…”

Section: Incremental Sfm With Distributed Camerasmentioning

confidence: 99%

“…1 While a generalized camera [17] may not explicitly include scale information, the camera model and accompanying pose methods [4,11,16] assume that the scale is known. 2 The "+++" is commonly used in literature to denote a non-degenerate rotation parameterization [9] 3. An improved solution method compared to [21].…”

mentioning

confidence: 99%

See 4 more Smart Citations

Large Scale SfM with the Distributed Camera Model

Sweeney

Fragoso

Höllerer

et al. 2016

2016 Fourth International Conference on 3D Vision (3DV)

View full text Add to dashboard Cite

We introduce the distributed camera model, a novel model for Structure-from-Motion (SfM). This model describes image observations in terms of light rays with ray origins and directions rather than pixels. As such, the proposed model is capable of describing a single camera or multiple cameras simultaneously as the collection of all light rays observed. We show how the distributed camera model is a generalization of the standard camera model and we describe a general formulation and solution to the absolute camera pose problem that works for standard or distributed cameras. The proposed method computes a solution that is up to 8 times more efficient and robust to rotation singularities in comparison with gDLS [21]. Finally, this method is used in an novel large-scale incremental SfM pipeline where distributed cameras are accurately and robustly merged together. This pipeline is a direct generalization of traditional incremental SfM; however, instead of incrementally adding one camera at a time to grow the reconstruction the reconstruction is grown by adding a distributed camera. Our pipeline produces highly accurate reconstructions efficiently by avoiding the need for many bundle adjustment iterations and is capable of computing a 3D model of Rome from over 15,000 images in just 22 minutes.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Numerical Stabilitymentioning

confidence: 99%

Section: Incremental Sfm With Distributed Camerasmentioning

confidence: 99%

Section: Incremental Sfm With Distributed Camerasmentioning

confidence: 99%

mentioning

confidence: 99%

See 3 more Smart Citations

Large Scale SfM with the Distributed Camera Model

Sweeney

Fragoso

Höllerer

et al. 2016

2016 Fourth International Conference on 3D Vision (3DV)

View full text Add to dashboard Cite

show abstract

Synchronization Problems in Computer Vision with Closed-Form Solutions

Arrigoni

Fusiello

2019

Int J Comput Vis

View full text Add to dashboard Cite

In this paper we survey and put in a common framework several works that have been developed in different contexts, all dealing with the same abstract problem, called synchronization by some authors, or averaging, or graph optimization by others. The problem consists in recovering some variables from a set of pairwise relation measurements. In particular, we concentrate on instances where the variables and the measures belong to a (semi-)group and the measures are their mutual differences (or ratios, depending on how the group operation is called). The groups we deal with have a matrix representation, which leads to an elegant theory and closed-form solutions.

show abstract

A center-driven image set partition algorithm for efficient structure from motion

Sun

Tao

2019

Information Sciences

View full text Add to dashboard Cite

Divide and Conquer: Efficient Large-Scale Structure from Motion Using Graph Partitioning

Cited by 24 publications

References 19 publications

Large Scale SfM with the Distributed Camera Model

Large Scale SfM with the Distributed Camera Model

Synchronization Problems in Computer Vision with Closed-Form Solutions

A center-driven image set partition algorithm for efficient structure from motion

Contact Info

Product

Resources

About