• Hardware Platform (Jetson / GPU) Nvidia RTX A2000 • DeepStream Version 6.0.1 • TensorRT Version 8.0.1-1 • NVIDIA GPU Driver Version (valid for GPU only) 470.86
Hello,
I have some question - I just want to process some frames in probe_fcn.
For instance I will get some bbox got from inference, want to crop the frame to retrieve some detected object, copy the object and process somehow - but not the whole frame - just an object.
And if in the probe_function I’m trying to deepcopy some frame - I got segmentation fault (core dumped).
I have prepared my own lib for gstlib but the problem also occurs in deepstream python example ( deepstream-imagedata-multistream
So is there any chance to get any information how it should be done?
It’s interesting cause as I have tested, it occurs only when I have uridecodebin/rtsp as source.
With usb/multifilesource it works ok.
The pipeline:
Yeah, I was thinking about this but why usb source works then?
And yeah, is there any possibility to process/deepcopy frames even when the source is rtsp?
Edit:
maybe it’s important - the problem also occurs in deepstream 5
All right, I have found the solution for the problem.
Dont know why but using Gst.Bin as sink to the streammux does not work.
Finally I have append the uridecodebin inside Gst.Bin, converters and caps filter for every stream are outside of GstBin.
You can take a look at the pipeline and see the differencess between the first pipeline I have added.
In not working version - the source pipeline (before streammux) was packed in GstBin.
You can take a look at pipeline.png pictures (remember to enlarge picture to see details).