Poor results tracking with sort or deep_sort.

[alexdominguez09]

Even I have managed to make work this project, I obtained similar results in terms of speed as it has been publish. YOLO with track Flag set to False : 30fps. In my case with a geforce 940M 2.24 fps

YOLO with track Flag set to True (deep_sort) : 14 fps. In my case with a geforce 940M 1.19 fps

My question is, the whole point of tracking, it is not only identify the same object within the different frames, but to accelerate the processes by not detecting with Yolo every single frame, so once in a while (periodically or when the detection is lost).

Then how come the speed with tracking is slower than the speed with only Yolo?.

And, any progress anyone in implementing the rest of the tracking algorithms from OpenCV (MIL,KCF,TLD,MEDIANFLOW) ?

thanks

[exotikh3]

@alexdominguez09 if I understood you point of view correctly you are asking why tracking is slower than running YOLO over each frame. Probably, answer is tracking task in this situation mostly about identifying position of specific object (person) at discrete period of time.

Optimizing is another tricky task. So if you want to do it please make sure you found a way to get bounding boxes of tracking objects without YOLO detector because "deep sort" based on metrics between encoding obtained from each detection on each frame.

[PBRAOS]

its too slow. with a GTX 1080 i get 12fps at best.

[bit-scientist]

Hi everyone, could someone actually solve the slow tracking problem. When I run, I am receiving fps of 10 and these kind of lines over and over

-cabac decode of qscale diff failed at 3 52
-error while decoding MB 3 52, bytestream 79098
-cabac decode of qscale diff failed at 129 82
-error while decoding MB 129 82, bytestream 32815
-cabac decode of qscale diff failed at 10 63`

thanks