Open CameronBodine opened 1 year ago
SRC correction sped up using math.sqrt
instead of np.sqrt
. Also used int()
instead of .astype(int)
. I see a speedup of 1.7x on the hello world
scan when processing EGN and exporting WCR rect/non-rect imagery! Great work @dbuscombe-usgs! Still unsure how to vectorize (if it's even possible) but considering that task as complete for the time being. May return if struck with divine numpy
inspiration in the future.
This will also speed up the substrate mapping workflow, including rectifying probs/logits since each class needs to be slant range corrected. Checking that task as complete also. Leaving issue open to identify other sources for speedup.
I found this beautiful solution to speeding up the warping process while rectifying sonar imagery:
https://github.com/scikit-image/scikit-image/issues/6864#issue-1653493908
For the example dataset using 7 threads for 7 images, the rectification process went from 23.8 seconds to 16.2 seconds! About to implement.
This issue will document workflows which would benefit from additional speed-up to lower overall processing time of sonar recordings.