yu-lab-vt / cinda Goto Github PK

Hello, what an impressive work, thank you for sharing this code!
in the process of calculating traj with traj, cost = mcc4mot (detection _ arcs, transition _ arcs), why are there some cases in which node ids do not have corresponding trajectories in the results of traj?

Thanks for publishing this offline tracking algorithm.

I think unique IDs in detection_arcs are enough to link detections as a trajectory. Do you have any suggestions to implement CINDA with unique IDs but NOT in the increasing order from 1 to n?
For example, I can have unique IDs in the following order 1, 200000000000, 3, 9, 5 and know how to put them in transition_mat. How can I run CINDA with this input?

1,3.467864,4.1917836,-6.7232538
200000000000,4.2795971,2.9235461,-7.1020659
3,1.7165976,2.4109024,-5.2415588
9,3.6710802,2.5107691,-5.7177933
5,3.3307807,3.359227,-6.3914796

I have tried to run CINDA with this input. It outputted an error (Error 5, wrong value of nodes). I do not know if I ignore (comment) this line (if ( tail < 0 || tail > n || head < 0 || head > n )) can possibly affect the output of CINDA.

It's a great project, but I have a problem running this project with my own data. If I have a data sequence and the detection results of the sequence, how can I create my own transition_ mat.txt and detecton_mat.txt.

WhenI try to run the notebook, I get the following error

lib_cinda_funcs.so is either not designed to run on Windows or it contains an error. Try installing the program again using the original installation media or contact your system administrator or the software vendor for support. Error status 0xc000012f

The returned pointer from pyCS2() function is never freed. I solved this problem by adding the following function to cinda_funcs.c :

void pyFreeTrackVec(price_t *track_vec) {
free(track_vec);
}

And then, in algo.py, at the end of mcc4mot():

_cinda.pyFreeTrackVec.argtypes = (ctypes.POINTER(ctypes.c_longlong), )
_cinda.pyFreeTrackVec.restype = None
_cinda.pyFreeTrackVec(track_vec)

return traj, cost

Awesome project! I've been doing research myself into same problem.

Would you be able to add a license to the code?

C_{i,j}, C_{en}, C_{ex} are well discussed in the paper (how to compute it) but I do not know how you can compute C_i

In your paper, C_i = log(beta_i / (1-beta_i). From your example code, C_i is in range [-8, -6] => beta_i is in range [0.00025, 0.0025]. This means the probability beta_i for detection x_i is approaching to zero.

How can I set beta_i or C_i for each detection in my dataset?