yu-lab-vt / mussp Goto Github PK

1. Thank Nuzhny007 to fix the memory leaks.
2. Change to C++14 compiler.
3. Fix the bug caused by 'abs' function under C++14.
4. Change the command of running muSSP.

I want to run this algorithm in an online manner i.e. I have new detections coming in and I want to grow the graph and optimize it at each time step. Is it possible with the current implementation?

Hi!
Thank you for this awesome work!
I'm a contributor to the open source project for multi-object tracking: https://github.com/Smorodov/Multitarget-tracker
And I want to integrate your awesome algorithm to this project. But in your code are some memory leaks (new without delete etc). I made some fixes in my fork: https://github.com/Nuzhny007/muSSP
Also was removed unused functions, fixed code style... And I can crate pull request with all fixes - what do you think about it?

I would like to try reproducing results, mentioned in the section 4.2 of the paper (Solving the min-cost flow approximation to the high-order modeling in MOT), but I am a bit vague on the exact steps, that need to be taken in order to do this.

From what I understood, 2 things are needed:

• Construction of graph with quadratic costs
• Optimization loop, which uses min-cost flow solver as a sub-routine

Can you please point to an appropriate publication with more details on this or an implementation, ideally the one that was used during preparation of the paper?

I checked in the main.cpp, it seems output will be written to file only when argc>3. However, there is no description of how to give these arguments.

@yu-lab-vt
Thank you for publishing so wonderful.
I'm a developper of OSS for automaus driving.
https://github.com/tier4/Pilot.Auto
I evaluated the muSSP and found that it was faster than ours and performed very well since the matrix size was 100x100.
I would really like to use this muSSP, but there is no description of the license.
Can you please add a license, Apache 2.0 would be great!

Hello, thanks you for sharing this code!
I have tried it and noticed that different trajectories may have some detections in common. Is this a wanted feature or am I doing something wrong?

Can you please give an example of how to write resulting trajectories to file?

According to the main.cpp, it seems that providing more than 2 command-line arguments should cause the call to print_solution(), and it indeed happens, but it doesn't write anything to file and the program terminates with a segmentation fault.

I've used provided file input_MOT_seq07_followme.txt as input, in case it helps.

Given certain graphs and edge weight magnitudes, Graph::extract_shortest_path() will fetch an invalid int for tmp_node_id, resulting in a segmentation fault. Example file is attached.
mussp_segfault_dimacs.txt.zip

Hello, I am very interested in your work. I would like to ask you about the input data, how the file input_MOT_seq07_followme.txt is generated