When using a primary detector with a secondary classifier the metadata from the classifier is sometimes lost, this seem to happen only at the very last frame of whatever input I send. I’ve tested this with different sinks, after debugging it looks like application is closing contexts to the classifier module before all data is returned.
This behaviour never happens when I run only the primary detector, is there anything I can change to force the pipeline to wait for the secondary before destroy?
• Hardware Platform x86 / Jetson AGX
• DeepStream 4.1 / 5
• TensorRT 6.0.1
• NVIDIA GPU Driver Version 418.56