faster motion correction

losonczylab / sima

Python package for analysis of dynamic fluorescence microscopy data

GNU General Public License v2.0

100 stars 50 forks source link

faster motion correction #8

Open pkaifosh opened 10 years ago

pkaifosh commented 10 years ago

Faster motion correction would be helpful when analyzing very large data sets. Some possible approaches include:

spatial down-sampling
algorithm changes
profiling and optimization of implementation
CPU parallelization
GPU support
cluster support

pkaifosh commented 9 years ago

Note that the current implementation of motion correction is about 4x slower than that in version 0.3

https://groups.google.com/forum/#!category-topic/sima-users/WLVCz5sdpjo

neurodroid commented 9 years ago

@llerussell is currently working on a motion correction implementation that is fast and performs well:

https://github.com/llerussell/sima/blob/dftreg/sima/motion/dftreg.py

I think that only some test cases and minor cleanups are missing for him to submit a pull request.

pkaifosh commented 9 years ago

Great! Adam mentioned this to me at codeneuro. It sounds like this may be a good improvement over our current cross-correlation method for whole frame alignment, and also for the first step of the HMM (which uses the whole-frame alignment to estimate parameters).