Large scale multi-view stereopsis evaluation

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 methodology. Although seminal, these benchmark datasets are limited in scope with few reference scenes. Here, we try to take these works a step further by proposing a new multi-view stereo dataset, which is an order of magnitude larger in number of scenes and with a significant increase in diversity. Specifically, we propose a dataset containing 80 scenes of large variability. Each scene consists of 49 or 64 accurate camera positions and reference structured light scans, all acquired by a 6-axis industrial robot. To apply this dataset we propose an extension of the evaluation protocol from the Middlebury evaluation, reflecting the more complex geometry of some of our scenes. The proposed dataset is used to evaluate the state of the art multiview stereo algorithms of Tola et al., Campbell et al. and Furukawa et al. Hereby we demonstrate the usability of the dataset as well as gain insight into the workings and challenges of multi-view stereopsis. Through these experiments we empirically validate some of the central hypotheses of multi-view stereopsis, as well as determining and reaffirming some of the central challenges.

Large-scale data for multiple-view stereopsis

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 different 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 tradeoff between the quality of the reconstructed 3D points (accuracy) and how much of an object’s surface is captured (completeness). Also, several 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 lack of texture and meshing (forming 3D points into closed triangulated surfaces).