The delay increases as the output video over time

nakayonum · June 21, 2022, 7:54am

Hi

Using jetpack 4.6’s “jetson-inferance” posenet.py
I want to process the input video 4ch.

Environment
・ Input video is delivered by RTSP, and the frame rate is 15fps.
Command

posenet.py --input-codec = h264 --network = resnet18-body rtsp: //192.1.10.152:12345

However, if you increase the resolution of the input video delivered by RTSP, the delay of the output video may or may not increase over time.
ex1): Delay does not increase for resolution 640x480
ex2): For a resolution of 1920x1280, the delay increases over time.

Why?

AastaLLL · June 21, 2022, 8:19am

Hi,

To narrow down the delay comes from, could you test the two experiments with the below sample as well?

github.com

dusty-nv/jetson-utils/blob/master/python/examples/video-viewer.py

#!/usr/bin/env python3
#
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Permission is hereby granted, free of charge, to any person obtaining a
# copy of this software and associated documentation files (the "Software"),
# to deal in the Software without restriction, including without limitation
# the rights to use, copy, modify, merge, publish, distribute, sublicense,
# and/or sell copies of the Software, and to permit persons to whom the
# Software is furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in
# all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER

This file has been truncated. show original

Thanks.

nakayonum · June 22, 2022, 6:48am

Hello Asta LLL

I did two tests with “video-viewer.py”.
The result was the same.
Since the result is the same for “video-viewer.py”, the processing speed of “videoSource.Capture ()” of “video-viewer.py” is slower than the frame rate delivered by rtsp, and the frame is “videoSource.Capture ()”. I think it’s in the buffer, what about?

dusty_nv · June 24, 2022, 12:48pm

If you are running this as 4x independent posenet.py processes, if you want to keep 1920x1080 resolution you may want to look to using DeepStream which is more optimized for handling multiple streams. As you have found, you could also try reducing the RTSP resolution, which often is okay since the DNN models run at a low resolution anyways (IIRC the pose estimation models are 300x300)

If you haven’t already, you could also try starting posenet with --input-rtsp-latency=0
It may also help adding rtpjitterbuffer element to the RTSP pipeline, I’m not entirely sure: https://github.com/dusty-nv/jetson-utils/blob/56d57f6713a5afe9029ab768ec5d13c285a52dfb/codec/gstDecoder.cpp#L528

nakayonum · July 1, 2022, 3:05am

Hi dusty_nv

Thanks for your comment
I will try “rtpjitterbuffer” as well.

system · July 15, 2022, 3:06am

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.