In the ssd objects detection algorithm,I need to find out the top 400(according to the paper) bouding boxes in the total 87732 anchor boxes.Therefore,we need a method to find out the maximum propabilities classes and it's corresponding boxes,and we call the propabilities as scores. np.argsort() may be a good function for that goal,it will return the sorted index from the smallest to the largest.Function and it's arguments list as follows:
- np.argsort(a,axis=-1,kind='quicksort',order=None)
a- a 1D array
axis- the dimention of sort.
The most important points about this function are:1.function will return the index of sorted array rather than arguments in the array.2.The sort order is from smallest to the largest.
There is an example to explain this function:
import numpy as np
a=np.array([1,4,2,4,6,6,2,8])
b=np.argsort(a,axis=-1)
print(b)Results:[0,2,6,1,3,4,5,7]
As we can see in the front examples,the minimum arguments in array a is 1 and it's index is 0,and the second smallest arguments is 2 in the index of 2 ... and so on.