Parallel load of vtp fiber bundles #144
Conversation
This version does the xml reading in threads, but updating the nodes is the time consuming part that still takes place in main thread.
Greatly improves loading performance
|
Hi Steve. It looks like this works great. Our most common use case right now is loading a MRML file with fiber bundles, where the MRML file describes the atlas and/or gives colors for the bundles according to some information like statistics we have generated with some external analysis. So the use case where we load an unannotated directory directly is not so common. I still think this feature is great, but we may not use it currently. I wonder if we should integrate this as a start point? And consider applying this to MRML files later? Or do we want to avoid adding this button and instead put this deeper into Slicer as a general feature for multiple vtk/vtp loads wherever they occur? |
|
@ljod yes, that's a good point. At this point this PR an experiment to see what we can do to split off some of the mrml loading from the main thread. Since it works this way we can think about how to generalize this to other storable mrml nodes. I suspect most could operate the same way but some may need work to be thread safe. Let me think about it. It would be best to do something that works for all Slicer data, but it might also be possible to do a mrml level solution that only works for dmri. |
|
@ljod or @zhangfanmark where is an example of a mrml file with lots of tracts of the type you mentioned above? |
This is ready for testing. The feature is a bit rough and we should discuss how to better integrate it with the gui and what features to add. E.g. it needs a progress bar.
Currently you select a directory filled with .vtp files and it does the file reading and some of the mrml operations in dedicated threads.
Here's a comparison of the traditional load on the right and the parallel load on the left:
https://www.dropbox.com/t/Iw1RoZmavrq2Y3bb