In this repo I vectorize IOU on multiple platforms and benchmark each in terms of runtime:
-
Numpy
-
Numba
-
Tensor Flow
Reading the csv into pandas
When box A's size is than 5 the non vectorized version performs better and when box A's size is greater than 5 numpy's vectorized version runs better
Tensor flow's runtime is pretty much constant all thru, and performs better when the number of boxes are higher, When the number of boxes are less the over head of copying the data to the GPU is greater than the processing time.
Another interesting observation here is that the for loop version of IOU performs better that the vectorized version when the number of boxes is greater than
