cascade feature extraction to detector

hi,
I am trying to do feature extraction after SSD,
so my pipline need to be:
SSD -> ParseBBox -> freature extraction

I am using deepstream-app and I have seen it have primary & secondary inference engine

my questions are:

  1. I have seen only examples of detector , can I run feature extraction network?
  2. its valid to do promary SSD with libnvdsinfer_custom_impl_ssd as plugin for parsing the BBox
    and those BBox will go to the secondary inference engine ?
  3. its possibale to send the features out of the pipe and not bbox or cropped images?

thanks
vadim

  1. I have seen only examples of detector , can I run feature extraction network?
    Test2 has the classifier example. You can replace it by your feature extraction network, set its output blob name.

  2. its valid to do promary SSD with libnvdsinfer_custom_impl_ssd as plugin for parsing the BBox
    and those BBox will go to the secondary inference engine ?

Yes. The metadata with all result is shared to primary, sencodary nvinfer.

  1. its possibale to send the features out of the pipe and not bbox or cropped images?

Yes. You can check the metadata type. If it can’t meet your need, you can create your custom metadata.