this still takes 100ms using the roi method
next step in this direction would need to make a cuda kernel which pipets the correct coordinates (e.g. with filtering on gpu-side) and returns just the results
this seems like a dead end (huge time investment for unlikely returns)
this still takes 100ms using the roi method next step in this direction would need to make a cuda kernel which pipets the correct coordinates (e.g. with filtering on gpu-side) and returns just the results
this seems like a dead end (huge time investment for unlikely returns)
closing..