What are the steps to be followed to filter out detections in the deepstream pipeline?

Let’s say in the current object detection plugin, immediately after object detection, there is tracker plugin that generates tracker id’s to the detected object be it any class, for eg person or car or any other object. I could remove the class detections by increasing the threshold, but let’s say in a single class person, I want to filter out detections such that only the filtered detections get assigned a tracker. How could I achieve this, I came across this because there was an option to filter trackers based on roi’s in the 360d parking systems, I wanted to try the same approach here too.


The simplest way is to hack the app with an unreachable threshold.

Ex. config_infer_primary.txt



Only top and bottom offsets are currently supported to filter out objects to be inferred. And you can modify the nvinfer plugin to add ROI