This is a pytorch implementation of Original code in numpy
I need to process large amount of XA videos. For a greyscale XA video of [34, 1, 512, 512] (Time,Channel,Height,Width), the original code will take 12 minutes by doing a for loop. I rewrite the code in pytorch with GPU support, now the same XA video takes 0.6 seconds.
The original code would take me a year to finish all my data, now it takes 2 days, 600 hundred times faster!
FrangiFilter2D(I)
Input: [B,1,H,W] the image to be filtered
Output: [B,1,H,W]