DeepStream 6.4 is out now supporting Ubuntu 22.04, JetPack 6.0 and GStreamer 1.20.3!
This release is packed with new features and improvements such as:
New Single-View 3D tracking capability available on NvTracker.
NvDCF tracker now supported with PVA backend on Jetson devices.
Support for AV1 encoder.
Two new plugins for image decoding and encoding.
And more!
We got the segment core dump when try uridecodebin source with setting protocols rtsp tcp (4).
we had tried to install the ffmpeg (apt-get install ffmpeg) inside docker deepstream v6.4 also give the segment core dump. the debug tool only show the exception above the av function ffmpeg av_opt_child_class_iterate coredump. I am not sure what is the main problem?
the command we tried: ffproble -rtsp_transport tcp -i "rtsp_stream_url"
There is an issue on using a .engine file inside a tracker ReID section.
In 6.3 config_tracker_xxx.yml, if providing onnxFile or uffFile, a corresponding engine file will be created, similar to model.onnx_b100_gpu0_fp32.engine. Then set path in modelEngineFile, the engine file can be used in future. For example:
ReID:
modelEngineFile: “/path/to/the/engine/file/converted/internally/or/externally.engine”
In 6.4, if providing a uff or onnx file, the experiment could run in the first time, and a .engine file will be create like previous, however, this .engine cannot be used while setting the modelEngineFile. Here is the error message:
[NvMultiObjectTracker] Loading TRT Engine for tracker ReID…
[NvMultiObjectTracker] Loading Complete!
Aborted (core dumped)
It seems the tracker ReID does not support a engine file properly in 6.4, but it always took some time to create a useless engine file in every call. I have tested the engine file in tracker ReID for both DeepSORT and DCF, both not working. This issue only related to engine file in ReID. Other engine models, for example, converted from onnx-file or provided by model-engine-file in a .txt like a detection model, still works fine.