Running 2 detection (Primary), with scaling between objects


(DS 4, Xavier)

I have the following setup:

  • first run object detection to detect “large” items (say using SSD, Yolo etc), collect all the objects found
  • for each object found, run another object detection to find “small” items, by scaling first the object.

For example (not the real use case, just to clarify) - first I detect all the dogs in the frame, then I take each dog detected, scale the image, and run detection for black spots, ticks and nails.

Any insights of how to build such a setup in efficient way in DS ?

Thanks !


It is workable.
For your use case, you will need two primary-gie and one ROI plugin.
Deepstream SDK use GStreamer frameworks and this allow you to compose your own pipeline efficiently.

It’s recommended to check our dsexample first.
The sample applies the detection and crop the ROI as the input of plugin, which can give you some idea.