Soft-NMS Gaussian Weights

Problem

Compute the Gaussian soft-NMS decay weights of candidate boxes relative to a kept box.

Each weight is exp(-IoU^2 / sigma), so overlapping neighbors keep a shrunken score instead of being deleted. Return the weights in order rounded to 4 decimal places; a non-positive sigma gives 0.0.

Test cases

case 1

in: [[0,0,2,2],[[0,0,2,2],[0,0,4,4],[10,0,12,2]],0.5]

expected: [0.1353,0.8825,1]

case 2

in: [[0,0,2,2],[[1,1,3,3],[0,0,1,1]],1]

expected: [0.9798,0.9394]

case 3

in: [[0,0,2,2],[],0.5]

expected: []

case 4

in: [[0,0,2,2],[[0,0,2,2]],0]

expected: [0]

Starter codePython

def soft_nms_gaussian_weights(box, boxes, sigma):
    # Your code here
    pass

4 test cases · run your code against them in the editor