Large-Scale Data for Multiple-View Stereopsis

Abstract

The seminal multiple-view stereo benchmark evaluations from Middlebury and by Strecha et al. have played a major role in propelling the development of multi-view stereopsis (MVS) methodology. The somewhat small size and variability of these data sets, however, limit their scope and the conclusions that can be derived from them. To facilitate further development within MVS, we here present a new and varied data set consisting of 80 scenes, seen from 49 or 64 accurate camera positions. This is accompanied by accurate structured light scans for reference and evaluation. In addition all images are taken under seven di↵erent lighting conditions. As a benchmark and to validate the use of our data set for obtaining reasonable and statistically significant findings about MVS, we have applied the three state-of-the-art MVS algorithms by Campbell et al., Furukawa et al., and Tola et al. to the data set. To do this we have extended the evaluation protocol from the Middlebury evaluation, necessitated by the more complex geometry of some of our scenes. The data set and accompanying evaluation framework are made freely available online.Based on this evaluation, we are able to observe several characteristics of state-of-the-art MVS, e.g. that there is a tradeo↵ between the quality of the reconstructed 3D points (accuracy) and how much of an object's surface is captured (completeness). Also, several H. Aanaes, R.R. Jensen and A.B. Dahl Technical University of Denmark, Lyngby, Denmark E-mail: {aanes, raje, abda}@dtu.dk G. Vogiatzis Aston University, Birmingham, England E-mail: g.vogiatzis@aston.ac.uk E. Tola Aurvis, Ankara, Turkey E-mail: tola@aurvis.com issues that we hypothesized would challenge MVS, such as specularities and changing lighting conditions did not pose serious problems. Our study finds that the two most pressing issues for MVS are those of meshing (forming 3D points into closed triangulated surfaces) and lack of texture.

A Critical Review of Automated Photogrammetric Processing of Large Datasets

“…Although recent benchmarks and software evaluations exist (Nex et al, 2015;Aanaes et al, 2016;Knapitsch et al, 2017;Schoeps et al, 2017), the paper focuses on more complex environments, surveyed with different platforms / cameras and comparison metrics are given following a standard terminology (Luhmann et al, 2014;Granshaw, 2016). …”

A Multi-View Stereo Algorithm Based on Homogeneous Direct Spatial Expansion with Improved Reconstruction Accuracy and Completeness

“…The data sets were from Technical University of Denmark (DTU) benchmarks [1,2], Middlebury benchmarks [3], or captured by ourselves. Figure 5 presents some typical images from the data sets used in the experiments.…”

“…Yet they were reconstructed successfully as shown in Figure 6. The temple and house model bore many regions with high curvature, e.g., the pillar in the red rectangle and the antenna (about one millimeter [1,2]) in a green rectangle in Figure 5. Yet all these highly-curved surfaces and thin structures were reconstructed clearly, as can be seen in Figure 6.…”

“…The parallax between any two selected images (including reference image) should be in the proper range; otherwise, matching accuracy might be reduced due to too small a parallax, and matching might even fail due to too few common regions caused by a large parallax. To ensure proper selection, scores for all candidate images can be calculated by Equation (1), and the image with highest score selected.…”

“…Table 1 provides a short list of such work, identified by their technical features. For more work in this field, we refer readers to the review articles presented in [1][2][3].…”