crossview_3d_pose_tracking icon indicating copy to clipboard operation
crossview_3d_pose_tracking copied to clipboard

Published 20 hours ago • verified publisher icon longcw

graph partitioning method for data association

Open linye-boli opened this issue 3 years ago • 6 comments

Hi, I found you refer to paper "Tracking Multiple People Online and in Real Time", which use "AL-ICM" method to solve correlation clustering problem, do you have the software which implements "AL-ICM" in python or other languages. I stuck on graph partition part

linye-boli avatar Sep 30 '20 06:09 linye-boli

Him I've wondered whether you solve correlation problem, stuck on this part either. Thanks

B10532021 avatar Dec 16 '20 03:12 B10532021

Hi guys, "Tracking Multiple People Online and in Real Time" used both binary integer programming solver and "AL-ICM" for tracking. They used to have a matlab version of the source code. Here is my ported BIP solver in python https://gist.github.com/longcw/654a86ffe11122079040a7615c99a627

It requires cvxopt and GLPK, hope it's helpful for you!

longcw avatar Dec 16 '20 03:12 longcw

Hi @longcw, Its definitely helpful! I'll take a look at it. Thank you very much!

Best Regards

B10532021 avatar Dec 16 '20 04:12 B10532021

Hey @longcw, I also have a question here. Could I ask why applying two kinds of matching, as Hungarian and graphpartition in the paper algorithm 1? Is the graphpartition just used for initialization like mvpose, then the following matching is Hungarian?

fandulu avatar Mar 12 '21 08:03 fandulu

@fandulu Yes, in the paper the graph partition is only for target creation. Hungarian is enough for the bipartite matching problem, but if you would like to use some different matching method that considering multiple frames or multiple views, or one person is split into multiple parts, then you may need the graph partition.

longcw avatar Mar 12 '21 10:03 longcw

@longcw， thanks for your answer!

fandulu avatar Mar 14 '21 03:03 fandulu