Help cropping output bounding boxes from nvinfer and sending them to a new nvinfer module

Accelerated Computing Intelligent Video Analytics DeepStream SDK
, ,
1

Hello!

We have a working pipeline that properly applies nvinfer to a camera source and performs tracking on the detection with nvtracker, and we print the bounding boxes and re-stream the video.

However, now we are interested in adding an element to the pipeline that, for each of the bounding boxes detected in the main camera source (depending on some criteria), generates a new cropped image and passes it through a classification model.

Is this possible using deepstream or would we have to develop a new custom callback that performs the crop and classification?

I’ve seen at this post that this was not possible in 2020, but I’m not sure about current state.

Regards

We are currently using this setup, but this is subject to change:
• Hardware Platform (Jetson / GPU) Currently testing in Ubuntu / GPU with a 3090 but will migrate to a Jetson Orin in the future
• DeepStream Version 6.0

3

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

Could you try to use the operate-on-gie-id and operate-on-class-ids to implement your needs?
https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_plugin_gst-nvinfer.html#id2