nvgstiva-app apply to yolo and webcam

thhsiao · May 6, 2020, 5:44pm

Hi,
My environment: Jetpack 3.3 (Cuda 9.0, TensorRT 4)+DeepStream1.5 for TX2

I run “nvgstiva-app -c ${HOME}/configs/PGIE-FP16-CarType-CarMake-CarColor.txt” and get correct result,
no matter what I config it for “1=CameraCSi” or “3=URI” in PGIE-FP16-CarType-CarMake-CarColor.txt.

But, when I run “nvgstiva-app -c config/nvgstiva-app_yolo_config.txt” in ~/deepstream-plugins-DS2,
only “3=URI” can be set in nvgstiva-app_yolo_config.txt and work correctly,
if “1=CameraCSi” was set in above file, the console show error messages:
------------> -----------------
** ERROR: <create_pipeline:1039>: Failed to link ‘multi_src_bin’ and ‘processing_bin_0’
** ERROR: <create_pipeline:1062>: create_pipeline failed
** ERROR: main:384: Failed to create pipeline
Quitting
App run failed

how to solve it for TX2 can run yolo to do the webcam objects detection under deepstream ?

AastaLLL · May 6, 2020, 5:44pm

Hi

Please use type=2 for the V4L2 camera.

You can find some information in our document:

Application Configuration
Source Group

<b>type</b>:<i> The type of the source. The other properties of the sourced depend on this type.</i>

1=Camera - NV Proprietary path (nvcamerasrc)
2=Camera - Open source V4L2 path (v4l2src) bypassing the ISP; this is suitable for use with USB cameras or CSI cameras that output in RGB or YUV formats.
3=URI - URI where an encoded stream is available (file/http/rtsp)

Thanks.

thhsiao · May 6, 2020, 5:44pm

Hi AastaLLL,

It is same result that use type=2 for the V4L2 camera.
Is there another modification in nvgstiva-app_yolo_config.txt?

same error message:
------------> -----------------
** ERROR: <create_pipeline:1039>: Failed to link ‘multi_src_bin’ and ‘processing_bin_0’
** ERROR: <create_pipeline:1062>: create_pipeline failed
** ERROR: main:384: Failed to create pipeline
Quitting
App run failed

AastaLLL · February 1, 2020, 7:27am

Okay.
Let me double check it first and update the information with you later.

AastaLLL · February 29, 2020, 1:24am

Hi,

We have checked this sample today and it works correctly on our side.

Please remember to re-compile the plugin component located at ‘sources/plugins/gst-yoloplugin-tegra’.
This will fix the error you shared in #3. You can find more information in our document:
https://github.com/vat-nvidia/deepstream-plugins/tree/DS2#note

Thanks.

thhsiao · February 1, 2020, 7:27am

Hi AastaLLL,

Thanks again, the situation is same as before even through I have re-build,
and this TX2 will be given back to another colleague, because our Xaivers are arrived.

Can you tell me which versions of Jetpack and deepstream should be installed to Xavier?
so that I can build deepstream-plugins to execute nvgstiva-app and yolo.

AastaLLL · February 1, 2020, 7:28am

Hi,

It’s weird since we can successfully execute it on our environment.

For Xavier, you will need DeepStreamSDK 3.0 and JetPack4.1.1:
https://developer.nvidia.com/embedded/deepstream-on-jetson-downloads

There are some required changes for Xavier and we plan to push it at the end of this week.
Thanks.

CJR · February 1, 2020, 7:28am

@thhsiao The git repo has been updated with changes required for DS3/Xavier.

thhsiao · February 1, 2020, 7:28am

Hi,
Thanks very much, this new git repo has include deepstream-plugins?
so that I need not download deepstream-plugins?
If I need download deepstream-plugins, it can be builded and executed on DeepStreamSDK 3.0 and JetPack4.1.1?

CJR · February 1, 2020, 7:29am

@thhsiao Its the same repo - https://github.com/vat-nvidia/deepstream-plugins which can now be used to run on Xavier / Jetpack 4.1 / DeepStream 3.0

shyoon · May 6, 2020, 5:44pm

I also have these symptoms in the same development environment.

gstyoloplugin.cpp
Line. 645
Changed to dsexample -> nvyolo

make && sudo make install

rm -rf ~/.cache/*
rm -rf ~/Model/*

nvidia@tegra-ubuntu:~/deepstream_reference_apps$ nvgstiva-app -c config/nvgstiva-app_yolo_config.txt
------------> -----------------
** ERROR: <create_pipeline:1039>: Failed to link ‘multi_src_bin’ and ‘processing_bin_0’
** ERROR: <create_pipeline:1062>: create_pipeline failed
** ERROR: main:384: Failed to create pipeline

[google drive code link]
https://drive.google.com/open?id=1xDzicN3yvaIfjkyMzaPLeBn82rrBaZzZ

[source0]
enable=1
#Type - 1=CameraCSi(none) 2=CameraV4L2 3=URI
type=2
camera-width=640
camera-height=480
camera-fps-n=30
camera-fps-d=1
camera-csi-sensor-id=0
camera-v4l2-dev-node=0
uri=file:///home/nvidia/sample_720p.mp4

Can you share a config for using yolov3-tiny in real time?

I did not know how to set app-mode.

0 = standalone; 1 = server; 2 = client

app-mode = 0

I need help.

beijing_PL · February 1, 2020, 7:30am

@AastaLLL Today,I got the same error with TX2. I have do it with ‘Tegra users working with Deepstream 1.5 and Jetpack 3.3 will have to regenerate the .cache files to use the standard caffe models available in the SDK. This can be done by deleting all the .cache files in /home/nvidia/Model directory and all its subdirectories and then running the nvgstiva-app using the default config file.’. But it didn’t work. I disable ds-example in cfg file.It can run ,just display. So can you help us to check it.

AastaLLL · February 1, 2020, 7:33am

Hi, beijing_PL

Sorry for the late reply.

Could you share what kind of error do you meet with us?
Thanks.

beijing_PL · May 6, 2020, 5:44pm

@AastaLLL after run ‘nvgstiva-app -c nvgstiva-app_yolo_config.txt’,print

------------> -----------------
** ERROR: <create_pipeline:1039>: Failed to link ‘multi_src_bin’ and ‘processing_bin_0’
** ERROR: <create_pipeline:1062>: create_pipeline failed
** ERROR: main:384: Failed to create pipeline
Quitting
App run failed

changed source0’s type from 3 to 2,run error

[source0]
enable=1
#Type - 1=CameraCSi 2=CameraV4L2 3=URI
type=2
camera-width=640
camera-height=480
camera-fps-n=30
camera-fps-d=1
camera-csi-sensor-id=0
camera-v4l2-dev-node=1
uri=file:///home/nvidia/sample_720p.mp4

AastaLLL · February 1, 2020, 7:33am

Hi,

We got a similar topic before. Could you check if this comment also works for you?
https://devtalk.nvidia.com/default/topic/1039068/deepstream-sdk-on-jetson/yolo-for-jetson-/post/5286880/#5286880

Thanks

kayccc · February 1, 2020, 7:37am

Hi beijing_PL,

Have you checked if previous suggestions works on your case?
https://devtalk.nvidia.com/default/topic/1039068/deepstream-sdk-on-jetson/yolo-for-jetson-/post/5286880/#5286880
Any result can be shared?

Thanks

guibli · May 6, 2020, 5:50pm

I’m also have this question,Running the nvgstiva-app [TEGRA platform only]

nvidia@tegra-ubuntu:~/TX2/tensorRT/deepstream_reference_apps-test$ sudo nvgstiva-app -c config/nvgstiva-app_yolo_config.txt
------------> -----------------
** ERROR: <create_pipeline:1039>: Failed to link ‘multi_src_bin’ and ‘processing_bin_0’
** ERROR: <create_pipeline:1062>: create_pipeline failed
** ERROR: main:384: Failed to create pipeline
Quitting
App run failed
nvidia@tegra-u

config file :
[source0]
enable=1
#Type - 1=CameraCSi 2=CameraV4L2 3=URI
type=2
camera-width=640
camera-height=480
camera-fps-n=30
camera-fps-d=1
camera-csi-sensor-id=0
camera-v4l2-dev-node=0
uri=file:///home/nvidia/TX2/tensorRT/football.mp4

environment:
DS 1.5
TX2
Jetpack 3.3

frame:
https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps/tree/7f27f22d674c64f4859dd1da896a43e35b0b9063

CJR · February 1, 2020, 8:17am

Hi,

DS 1.5 is an old release please wait for DS4.0 which will be released soon.

guibli · February 1, 2020, 8:17am

Thanks for you answer ,DS4.0 will also support TX2 platform,right?

CJR · February 1, 2020, 8:17am

yes