How to do to run feature extraction model , for eg. face recognition in deepstream python apps

Hi, I am new to deepstream framework and have no prior experience of working with CPP, I am trying to develop a face recognition application for CCTVS for which I need to learn to customize deepstream python sample apps. Sharing of resources would be helpful …

• Hardware Platform ( GPU)–> Tesla k40m
• DeepStream Version --> 5.0
• NVIDIA GPU Driver Version --> 10.2
• Issue Type -->question

Deepstream SDK is based on gstreamer, so before start with deepstream, you need to learn gstreamer first. Otherwise, it will be very hard to understand the sample codes and customize the deepstream app.

It is also necessary to read and understand the user documents of deepstream SDK

DeepStream SDK is c/c++ based while it also provides python bindings to support python app development, before that it is recommended to know some basic knowledge of gst-python.,,

After all these preparation, you can start with our sample python apps.

The developer of deepstream python app should have C/C++ experience especially when he needs to customize the apps.

Thankyou Fiona Chen for your reply, much appreciated. Still I need to know if we can customize plugins to do feature extraction other than classification and detection. And if there is any app that had done face recognition kindly share.

There are four predefined network type : Detector, Classifier, Segmentation, InstanceSegmentation. If your network is none of the four types, it should be the other type(type value is 100). The enum NvDsInferNetworkType definition is in /opt/nvidia/deepstream/deepstream-5.0/sources/includes/nvdsinfer_context.h

When use other type network, the related object meta, classifier meta data, … will be useless for the network You need to define your own meta data to be used by downstream elements in the pipeline.

Due to law, we can not provide face recognition implementation.

Thanks for all the help…