OD as secondary detector in Python

[monita.ramirezb]

• Hardware Platform: Jetson Xavier NX
• DeepStream Version: 5.0
• Language: Python
I’m working with Python Apps (deepstream_test_2) because I need to run a secondary detector. Can someone point me on how to:

• Access information of the sgie (For example, in test_2, how to access the color or make of the car) to change the label above the bounding box.
• Configure my sgie as a detector instead of a classifier (I already trained my model in TLT and have my .etlt file)

I’ve seen many comments on a “back-to-back model”, but I need this on Python. How else can this be done?

[Amycao]

@monita.ramirezb

• Access information of the sgie (For example, in test_2, how to access the color or make of the car) to change the label above the bounding box.

→ Is it easy to modify the labels in labels file? infer postprocessing will parse the labels files and attach it as user meta data.
sources/libs/nvdsinfer/nvdsinfer_context_impl.cpp InferPostprocessor::parseLabelsFile

• Configure my sgie as a detector instead of a classifier (I already trained my model in TLT and have my .etlt file)

→ You can not set classifier as detector, since the post processing is different.

See inline.

[monita.ramirezb]

@monita.ramirezb

• Configure my sgie as a detector instead of a classifier (I already trained my model in TLT and have my .etlt file)

→ You can not set classifier as detector, since the post processing is different.

Oh, I understand.
Then, instead of test_2, what would be the process to implement, for example, a licence plate detector in Python ?

[Amycao]

You can refer to back to back detector, deepstream_reference_apps/back-to-back-detectors at master · NVIDIA-AI-IOT/deepstream_reference_apps · GitHub, it’s for c version, you need to implement using python.