Manually add Metadata to replace PGIE object detector

[paul.doucet]

Please provide complete information as applicable to your setup.

• Jetson Xavier
• DeepStream 6.0.0
• JetPack 4.6.2
• TensorRT 8.0.1
**• Questions **

Hi,
I need to use an external object detector instead of the PGIE object detector, i.e. I will receive object coordinates from an outside app and parse them. Now I need to somehow feed these object coordinates as input to the nv-infer classifier module. In my understanding I need to add a probe upstream and use nvds_add_obj_meta_to_frame to manually add the metadatas, does it seem feasible to you with multiple input streams and multiple object? Has this ever be done before?

Thank you very much,

Paul Doucet

[yuweiw]

Sorry for late response. You can try to use the Gst-nvdsvideotemplate plugin to implement your needs.
nvdsvideotemplate.
There are some demos for this: tao_others.

[paul.doucet]

Thanks, in the end I implemented this by putting a probe on the SGIE sink and by manually adding the objects with nvds_add_obj_meta_to_frame