Cycle consistency filtering writes plots in the run() method, without any dask.annotate() call to specify which machine to write these plots on. This fails on the cluster because there is no guarantee that the directory to write the plots to will exist on the machine running cycle consistency. Even if there is, the plots will not be obtained on the machine from which gtsfm is run.
akshay-krishnan
commented
1 year ago
Cycle consistency filtering writes plots in the run() method, without any dask.annotate() call to specify which machine to write these plots on. This fails on the cluster because there is no guarantee that the directory to write the plots to will exist on the machine running cycle consistency. Even if there is, the plots will not be obtained on the machine from which gtsfm is run.
https://github.com/borglab/gtsfm/blob/4a9dc6a7ae9a1146b3fc056827b265776921b49c/gtsfm/view_graph_estimator/cycle_consistent_rotation_estimator.py#L137