Object Re-Identification


I want to ask if there’s a possible solution to do object re-identification using deepstream pipeline.
The idea is combining multistreamtiler plugin and tracker plugin of the pipeline to identify the same object from two different sources.
I was thinking to perform pose estimation using a second inference plugin that could generate modified metadata but a bit stuck on how to include it in deepstream pipeline.
Any idea will be really helpful?


Hi ,
Please provide your setup info as other topics.

You can refer to https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps/tree/master/back-to-back-detectors , if the 2nd gie is the same as the 1st gie, it should be what you want - object re-identification.

Hello guys,

  1. If I’m not mistaken, we can implement the following pipeline:
    src1 ->
    -> multistreamtiler -> nvinfer_1 -> nvinfer_2 -> nvtracker -> …etc…
    src2 ->

But how to match bbox from the first nvinfer to bbox from the second nvinfer and check that it’s the same objects ?

  1. Could you explain what does back-to-back detection mean?

  2. Actually, how to use two stream sources and match objects between them?

Hi rostislav.etc,

Please help to open a new topic for your issue. Thanks