Where to get values for NvDsInferParseDetectionParams

[mfoglio]

I am writing my own parser for an object detection model. I am using C++, DeepStream 6.2 and a Tesla T4.
I see that I can get the following values from NvDsInferParseDetectionParams: numClassesConfigured. perClassPreclusterThreshold, and perClassPostclusterThreshold (see NVIDIA DeepStream SDK API Reference: NvDsInferParseDetectionParams Struct Reference ).
Where are these values read from? Are they specified in primary engine config txt file as in the example below?

[property]
num-detected-classes

[class-attrs-all]
pre-cluster-threshold=0.3
post-cluster-threshold=0.

If the nms-iou-threshold is specified in [class-attrs-all], is there a way to read its value from the C++ code?

[yuweiw]

@mfoglio

Where are these values read from? Are they specified in primary engine config txt file as in the example below?

They were parsed from the configuration file. Please refer to our source code gstnvinfer_property_parser.cpp and gstnvinfer_yaml_parser.cppin the opt\nvidia\deepstream\deepstream\sources\gst-plugins\gst-nvinfer path.

[mfoglio]

Hi @yuweiw , thank you for your reply.
I understand that numClassesConfigured , perClassPreclusterThreshold , and perClassPostclusterThreshold are parsed from the pgie txt configuration file.
Is there a way I can also get the nms-iou-threshold value in my custom C++ postprocessing for object detection?

[yuweiw]

Yes. Please refer the demo code nvdsinfer_custombboxparser_tao.cpp.

[mfoglio]

It looks like it’s not possible to retrieve nms-iou-threshold, at least looking at that file. Thank you anyway.

[yuweiw]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

There are many apps of this post-processing function in the following demo: deepstream_tao_apps. You can refer to that too.