the attempt to use the ants motion_correction workflow is problematic. it doesn't seem possible to successfully save the transform files in a location other than /tmp, and this can be problematic on HPC systems since tmp can run out of space.
the current model does the full run on channel 1 and then separately transforms channel 2, which requires holding onto all of teh transforms. I propose to change to a model where the transform is computed for each timepoint, and then applied, so that the transform files can then be deleted before the next step.
poldrack
commented
2 years ago
the attempt to use the ants motion_correction workflow is problematic. it doesn't seem possible to successfully save the transform files in a location other than /tmp, and this can be problematic on HPC systems since tmp can run out of space.
the current model does the full run on channel 1 and then separately transforms channel 2, which requires holding onto all of teh transforms. I propose to change to a model where the transform is computed for each timepoint, and then applied, so that the transform files can then be deleted before the next step.