Distributed and Graph-based Structure from Motion. This project includes the official implementation of our Pattern Recognition 2020 paper: Graph-Based Parallel Large Scale Structure from Motion.
Using smart pointer to avoid copying of reconstructions and database caches in distributed sfm stage.
The maximum virtual memory usage drops from 403GB to 245GB (40 percent) on master node when running distributed sfm on a dataset which contains 100 thousand images.
Besides, avoid copying of large chunk of memory saves roughly 1.5 days in my case(120GB of RAM, 400GB of swap area).
Using smart pointer to avoid copying of reconstructions and database caches in distributed sfm stage. The maximum virtual memory usage drops from 403GB to 245GB (40 percent) on master node when running distributed sfm on a dataset which contains 100 thousand images. Besides, avoid copying of large chunk of memory saves roughly 1.5 days in my case(120GB of RAM, 400GB of swap area).