General question about tracker followed by detector

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

• Hardware Platform (Jetson / GPU) : jetson
• DeepStream Version : 5.0 GA
• JetPack Version (valid for Jetson only) :4.4
• TensorRT Version:7.1.X

A s you know, the deepstrem has detector and trackers, and for detector nvinfer plugin runs and it has interval property, and this property skip the processing of frame as long interval. the tracker followed by detector, I want to know, when I set interval to 0, the detector has to process all of frames and detector localize the all of objects in frames, so what’s role of tracker here? only assign ID for boxes? Or refine the boxes and assign a ID ?

3

If you use NvDCF tracker, it will refine the bbox and assign tracker Id

4

The tracker-width and tracker-height are region of processing surrounding around detected objects? or the whole of input image resized into these size?

5

Whole image

6

@bcao,
The idea of tracking is that instead of processing of whole image like detector, the tracking have to process the surrounding around detected objects instead of whole image to have less computation than detector.

7

Tracker also do tracking based on the objects， if we resize the image to a small resolution, then it will reduce the computation.