Detected objects are lost in probe

I try to make an object detection application with python3 which based on Nvidia deepstream 6.0, the plugins from Nvidia – nvinfer, nvtracker, nvdsanalytics, are used as the inference, tracker and analystics elements.

The pipeline likes:
RTSP source → streammux → queue → infer → queue → tracker → queue → analytics → queue → video convert → queue → caps filter → queue → tiler → queue → nvosd → queue → video convert → caps filter → encoder → h264 parser → splitmux

A data probe is added to the sink pad of tiler, in this probe, the meta data is filtered out, and the user meta data is extracted from meta data, and then, the detected objects are filtered from the user meta data.

At the sink end of pipeline, a video is recorded by splitmux with the detected objects which marked with rectangle by nvosd plugin.

Unfortunately, some detected objects which marked on the video are lost, these detected objects could not be filtered out in the probe, this leads the app could not get all the detected objects.

The probe is tried to be added to the source pad of analytics , sink pad of queue which linked between the analytics and video convert, or the sink pad of nvosd, the source pad of nvosd, the results are same with the probe is added to the sink pad of tiler.

What could be the possible reason about the dected objects are lost in the probe in my python application?



Could you share a simple source code so we can reproduce internally first?

Please check the attached python source code. (19.0 KB)

With a senior engineer support, I am told the possible problem reason is, the metadata of detected objects is extracted in the loop of user meta list in the attached python code, but nvosd extracts the metadata of detected objects in the loop of object meta list, not user meta list, thus, the app misses the objects which has no user meta data.

I wonder the reason that almost object meta has user meta data, but some object meta has not user meta data, actually, there is no user meta data attached to object meta by the application, it seems the analytics generates the user meta data and attached to object meta.

After the metadata of detected objects is extracted in the loop of object meta list, instead of user meta list, the problem goes away.

I’m closing this topic due to there is no update from you for a period, assuming this issue was resolved.
If still need the support, please open a new topic. Thanks


It’s this issue fixed by extracting the objects in the meta list within the loop?