#hungarian algorithm for torch#
Wrapping c implementation of Hungarian algorithm (or Munkres assignment algorithm) to torch/lua
The c code comes from Cyrill Stachniss's work
Compute 1000x1200 matrix only takes 0.28s with Intel 4790K
luarocks install hungarian
require 'hungarian'
-- input must be torch.IntTensor
matrix = torch.IntTensor({{10,19, 8,15},
{10,18, 7,17},
{13,16, 9,14},
{12,19, 8,18},
{14,17,10,19}})
min_assignment = hungarian.minCost(matrix)
max_assignment = hungarian.maxCost(matrix)Also see test.lua for example
th test.lua
Expected results:
test1 input:
10 19 8 15
10 18 7 17
13 16 9 14
12 19 8 18
14 17 10 19
[torch.IntTensor of size 5x4]
test1 output:
1
3
4
0
2
[torch.IntTensor of size 5]
test1 rearrange result:
10 19 8 15
14 17 10 19
10 18 7 17
13 16 9 14
[torch.IntTensor of size 4x4]
test2 input:
10 10 13 12 14
19 18 16 19 17
8 7 9 8 10
15 17 14 18 19
[torch.IntTensor of size 4x5]
test2 output:
1
5
2
3
[torch.IntTensor of size 4]
test2 rearrange result:
10 10 13 12 14
8 7 9 8 10
15 17 14 18 19
0 0 0 0 0
19 18 16 19 17
[torch.IntTensor of size 5x5]
TEST PASS!