something miss with sort and match?

[nickbp760]

a = torch.tensor([[0.9, 0.01],[0.8,0.02],[-2.3, -2.4], [-2.4, -1.5]]) #gallery print(a.size()) b = torch.tensor([[2.1, 0.4],[-2.1,-3.2],[-2.2,-0.3]]) #query print(b.size()) dismat = torch.cdist(b, a) print(dismat) m,n=dismat.shape

Result torch.Size([4, 2]) torch.Size([3, 2]) tensor([[1.2618, 1.3544, 5.2154, 4.8847], [4.3936, 4.3334, 0.8246, 1.7263], [3.1155, 3.0170, 2.1024, 1.2166]]) torch.Size([3, 4])

import numpy as np gallery_ids= np.array([1,1,2,2]) #gallery query_ids = np.array([1,2,2]) #query

print(n[:, np.newaxis]) indices=np.argsort(dismat,axis=1) print(indices) print(n[indices]) matches = (gallery_ids[indices] == query_ids[:, np.newaxis]) print(matches)

Result [[1] [2] [2]] tensor([[0, 1, 3, 2], [2, 3, 1, 0], [3, 2, 1, 0]]) [[1 1 2 2] [2 2 1 1] [2 2 1 1]] [[ True True False False] [ True True False False] [ True True False False]]

It's strange when your algorithm detect that true for index [1][0],[1][1] and index [2][0],[2][1]? because the euclidan said that minimum distance is [1][2],[1][3] and index [2][2],[2][3] like my index that I made i just want to have an idea about your code in

Sort and find correct matches

indices = np.argsort(distmat, axis=1)
matches = (gallery_ids[indices] == query_ids[:, np.newaxis]) 

[nickbp760]

finally i understood that [[1 1 2 2] [2 2 1 1] [2 2 1 1]] it's the prediction