The backprojection ought to be the slower of BP/FP - can you argue why this is?
Right now our BP_test_2 is faster than our FP.
Can you use the ideas from the BP_test_2 algorithm to accelerate FP?
The core ideas are:
Partition the problem into flat and steep angles in a preprocessing step.
Do as few square root calculations as possible.
Test the two improvements individually and try to argue why they give/don't give a speedup.
The backprojection ought to be the slower of BP/FP - can you argue why this is? Right now our BP_test_2 is faster than our FP. Can you use the ideas from the BP_test_2 algorithm to accelerate FP?
The core ideas are:
Test the two improvements individually and try to argue why they give/don't give a speedup.![comparison of runtimes](https://user-images.githubusercontent.com/41566360/53304362-9c3b3b00-3874-11e9-871d-da659ead7c85.png)