Memory issue in video analytics pipeline with dlstreamer elements

[AI_researcher]

I am implementing the computer vision use case for object detection with Dlstreamer pipeline.

but, facing memory related issue “memory is increasing exponentially while running pipeline”.

Running the Manual pipeline in latest dlstreamer container with 3 Dlstreamer elements:
1. Gvadetect, 2. Gvaclassify, 3. Gvawatermark,  found container memory is increasing exponentially.

Analytics Pipeline:
gst-launch-1.0 filesrc location=Test_input_Video1.mp4 ! decodebin ! videoscale n-threads = 40 ! video/x-raw,width=832,height=468 ! videoconvert name=\"videoconvert\" n-threads=40 ! video/x-raw,format=BGR ! queue leaky=upstream ! gvadetect ie-config=CPU_BIND_THREAD=NO model="/home/person-detection-0201/1/FP32/person-detection-0201.xml" model-proc="/home/person-detection-0201/1/person-detection-0202.json" threshold=0.5 name="detection" inference-interval=1 ! queue ! gvaclassify inference-region=roi-list model="/home/classification/saved_model.xml" model-proc="/home/classification/1/Model_proc.json" ! gvawatermark ! queue ! videoconvert ! video/x-raw,format=I420 ! avenc_mpeg4 bitrate=2000000 ! mpeg4videoparse ! mux.video_0 qtmux name=mux ! filesink location=Test_output_video1.mp4

Please help to resolve memory issue. 

[Iffa_Intel]

Hi,

Could you clarify:

1. Which OpenVINO sample application/code are you using?
2. Which model are you using for this use case? (if possible, please help to share details of model(eg custom model/pretrained model), relevant model files and conversion steps that you did)
3. Your OS and hardware details

Cordially,

Iffa

[Iffa_Intel]

Hi,

Thank you for your question.  If you need any additional information from Intel, please submit a new question as Intel is no longer monitoring this thread.

Cordially,

Iffa