| Popis: |
Photogrammetry is used to build 3-dimensional models of everything from terrains to ancient statues. In the past, the stitching process was done on powerful PCs and could take weeks for large datasets. Even relatively small objects often required several hours to stitch together. With the availability of parallel processing options in the latest release of Agisoft PhotoScan, it is possible to leverage the power of high performance computing on large datasets.This poster presents the results of benchmarking tests for three datasets processed at two different model quality levels, medium and high (there are four times more points in the dense point cloud at the high setting) using 2, 4, 8, and 16 nodes. The purpose of the benchmarking is to determine how to optimize software license usage and compute resources against time and output quality.The poster also details the matrix of user-specified parameters that have been built into the python script that submits the parallel jobs. These parameters have evolved through the assessment of needs of users who are using the HPC deployment of photogrammetry as a service. We are excited by the uptake of this new service around campus in different fields across multiple disciplines. A group of cultural heritage experts will be using the service from Italy this summer, and a graduate student in anthropology will be stitching aerial data sets from Mexico. |
