Detail of parameters of DeepSORT, on DeepStream6.0

Hello.

**• Hardware Platform (Jetson / GPU): jetson
**• DeepStream Version: 6.0
**• JetPack Version (valid for Jetson only): 4.6
**• Issue Type( questions, new requirements, bugs): questions

I have been trying to use DeepSORT with NvMultiObjectTracker, pre-implemented library, on Deepstream 6.0.
And I checked out the default yml file of it provided by deepstream and its document on web site.

But I can’t understand some parameters.
So I would like to know about the detail of its parameters, especially below.

First,

  • stateEstimatorType
  • reidType
    For these params, I understood DUMMY means that the algorithm won’t be used to track objects.
    For example, if I set stateEstimatorType=0, only ReID will be used. Is this correct?

Second,

  • reidHistorySize
    If frames of the number of set size passed, the oldest feature vector of each target tracker is deleted even if the frame that a detector can’t find any object is included in these frames?
    For example, if I set this to 2 and 3 frames have passed since tracking started, first feature vector of each tracker object will be deleted. But how about the situation where a detector can’t find any object at a frame out of 3 frames? This means each tracker object has only 2 features after passing 3 frames. In this situation, the features of each object won’t be erased. Is this correct?

Thank you,

Below is what I can share to help you understand it. Hope you can change those parameters and have a try the impact of those parameters.

Gst-nvtracker — DeepStream 6.1 Release documentation (nvidia.com)

/opt/nvidia/deepstream/deepstream-6.1/samples/configs/deepstream-app/config_tracker_DeepSORT.yml

@kesong

Thank you for your reply.

I am reading the document but I can’t figure out in some ways.
Can I ask you if I set stateEstimatorType to 0, which means “DUMMY”, how this algorithm calculate mahalanobis distance?

Thank you.

There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.
Thanks

SIMPLE and REGULAR use Kalman Filter for state estimation. DUMMY will do easy state update.

stateEstimatorType: 2 # the type of state estimator among { DUMMY=0, SIMPLE=1, REGULAR=2 }

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.