Yolo for Jetson

CJR · August 31, 2018, 9:35pm

A sample standalone app for running inference using yolo in jetson is now available here https://github.com/vat-nvidia/deepstream-plugins

Tegra users who are currently using DeepStream 1.5 can use this sample app as a reference and incorporate a similar pipeline in their inference plugins.

albertr · September 4, 2018, 4:50pm

Nice! Thanks for sharing.

-albertr

prince15046 · September 12, 2018, 5:57am

While making it in Jetson TX2.
I am facing below error.

Can you point out what is the problem.

/home/nvidia/YOLOTEST/deepstream-plugins/sources/gst-yoloplugin/yoloplugin_lib

After make.

yolo.cpp:310:25: error: ‘class nvinfer1::IConcatenationLayer’ has no member named ‘setAxis’
concat->setAxis(0);

CJR · September 13, 2018, 4:43am

@prince15046 Which version of Jetpack are you currently using ?

prince15046 · September 14, 2018, 1:50pm

Following the below thread now -

https://devtalk.nvidia.com/default/topic/1038913/deepstream-sdk-on-jetson/deepstream-plugins-make-failed/post/5281057/?offset=14#5281547

Can you help in building the TRT-yolo. Facing the following error.

../../gst-yoloplugin/yoloplugin_lib/build/yoloplugin_lib.o: In function `YoloPluginProcess':
yoloplugin_lib.cpp:(.text+0xf00): undefined reference to `cv::dnn::experimental_dnn_v3::blobFromImages(std::vector<cv::Mat, std::allocator<cv::Mat> > const&, double, cv::Size_<int>, cv::Scalar_<double> const&, bool, bool)'
collect2: error: ld returned 1 exit status
Makefile:49: recipe for target 'TRT-yolo-app' failed
make: *** [TRT-yolo-app] Error 1

subhasis.chakraborty · September 21, 2018, 5:02am

Facing this error while compiling the libgstyolo in Jetson-TX1, with deep stream-2.0 for running the TRT-yolo app in the Jetson TX-1 device

/usr/bin/ld: skipping incompatible /usr/local/deepstream/libgstnvquery.so when searching for -lgstnvquery
/usr/bin/ld: cannot find -lgstnvquery
/usr/bin/ld: skipping incompatible /usr/local/deepstream/libgstnvdsmeta.so when searching for -lgstnvdsmeta
/usr/bin/ld: cannot find -lgstnvdsmeta
collect2: error: ld returned 1 exit status
Makefile:64: recipe for target ‘libgstnvyolo.so’ failed
make: *** [libgstnvyolo.so] Error 1

prashanth.bhat · September 21, 2018, 6:15am

Hi NvCJR:

Sorry, I’m a bit confused about the environment where this new YOLO plugin can be used. The README file says it’s for Deepstream 2.0. Your post says that it’s for Jetson. However, Deepstream 2.0 is not yet available on the Jetson platform. hence my confusion.

Please clarify.

Thanks.

prashanth.bhat · September 21, 2018, 6:17am

Another question:

In README.md:
Add absolute paths of images to be used for calibration in the calibration_images.txt file within the sources/gst-yoloplugin/yoloplugin_lib/data directory.

Could you elaborate on the required format of images for calibration? Sorry if this is a stupid question. Thanks.

subhasis.chakraborty · September 25, 2018, 2:32pm

Getting this error while compiling TRT-yolo app on Jetson-TX2

trt_utils.cpp: In function ‘nvinfer1::ILayer* netAddUpsample(int, std::map<std::__cxx11::basic_string<char>, std::__cxx11::basic_string<char> >&, std::vector<float>&, int&, nvinfer1::ITensor*, nvinfer1::INetworkDefinition*)’:
trt_utils.cpp:584:5: error: ‘IConstantLayer’ is not a member of ‘nvinfer1’
     nvinfer1::IConstantLayer* preM = network->addConstant(preDims, pre);
     ^
trt_utils.cpp:584:31: error: ‘preM’ was not declared in this scope
     nvinfer1::IConstantLayer* preM = network->addConstant(preDims, pre);
                               ^
trt_utils.cpp:584:47: error: ‘class nvinfer1::INetworkDefinition’ has no member named ‘addConstant’
     nvinfer1::IConstantLayer* preM = network->addConstant(preDims, pre);
                                               ^
trt_utils.cpp:611:5: error: ‘IConstantLayer’ is not a member of ‘nvinfer1’
     nvinfer1::IConstantLayer* post_m = network->addConstant(postDims, post);
     ^
trt_utils.cpp:611:31: error: ‘post_m’ was not declared in this scope
     nvinfer1::IConstantLayer* post_m = network->addConstant(postDims, post);
                               ^
trt_utils.cpp:611:49: error: ‘class nvinfer1::INetworkDefinition’ has no member named ‘addConstant’
     nvinfer1::IConstantLayer* post_m = network->addConstant(postDims, post);
                                                 ^
trt_utils.cpp:616:5: error: ‘IMatrixMultiplyLayer’ is not a member of ‘nvinfer1’
     nvinfer1::IMatrixMultiplyLayer* mm1
     ^
trt_utils.cpp:616:37: error: ‘mm1’ was not declared in this scope
     nvinfer1::IMatrixMultiplyLayer* mm1
                                     ^
trt_utils.cpp:617:20: error: ‘class nvinfer1::INetworkDefinition’ has no member named ‘addMatrixMultiply’
         = network->addMatrixMultiply(*preM->getOutput(0), false, *input, false)
                    ^
trt_utils.cpp:621:5: error: ‘IMatrixMultiplyLayer’ is not a member of ‘nvinfer1’
     nvinfer1::IMatrixMultiplyLayer* mm2
     ^
trt_utils.cpp:621:37: error: ‘mm2’ was not declared in this scope
     nvinfer1::IMatrixMultiplyLayer* mm2
                                     ^
trt_utils.cpp:622:20: error: ‘class nvinfer1::INetworkDefinition’ has no member named ‘addMatrixMultiply’
         = network->addMatrixMultiply(*mm1->getOutput(0), false, *post_m->getOut
                    ^
Makefile:62: recipe for target 'build/trt_utils.o' failed
make[1]: *** [build/trt_utils.o] Error 1
make[1]: Leaving directory '/home/nvidia/Desktop/deepstream-plugins/sources/gst-yoloplugin/yoloplugin_lib'
Makefile:68: recipe for target 'yoloplugin_lib/libyoloplugin.a' failed
make: *** [yoloplugin_lib/libyoloplugin.a] Error 2

CJR · September 25, 2018, 7:15pm

@subhasis.chakraborty Which jetpack are you using ?

subhasis.chakraborty · September 26, 2018, 12:59am

@NvCJR

I ma using Jetpack 3.3

CJR · September 26, 2018, 2:47am

@subhasis.chakraborty TensorRT offcially supports constant and matrixmultiply layers since 4.0.1 release. Can you reinstall tensor RT using the latest package of Jetpack 3.3 and check if that resolves your issue? It’s available here - https://developer.nvidia.com/embedded/downloads#?search=jetpack%203.3

prince15046 · September 26, 2018, 2:42pm

Hi NvCJR,

The plugin works really good.

Is there any plan to work on Python plugin for the same ?

CJR · September 26, 2018, 7:03pm

@prince15046 we dont have a python version of the plugin but gstreamer offers python bindings which may be helpful to you. Reference - https://gstreamer.freedesktop.org/bindings/python.html

albertr · September 27, 2018, 11:26pm

g++ -I"/include" -I"/usr/local/cuda-9.0/include" -I “/usr/local/include” -c -o build/trt_utils.o -O2 -std=c++11 -lstdc++fs -fPIC -Wall -Wunused-function -Wunused-variable pkg-config --cflags glib-2.0 trt_utils.cpp
trt_utils.cpp: In function ‘nvinfer1::ILayer* netAddUpsample(int, std::map<std::__cxx11::basic_string, std::__cxx11::basic_string >&, std::vector&, int&, nvinfer1::ITensor*, nvinfer1::INetworkDefinition*)’:
trt_utils.cpp:584:5: error: ‘IConstantLayer’ is not a member of ‘nvinfer1’
nvinfer1::IConstantLayer* preM = network->addConstant(preDims, pre);
^
trt_utils.cpp:584:31: error: ‘preM’ was not declared in this scope
nvinfer1::IConstantLayer* preM = network->addConstant(preDims, pre);
^
trt_utils.cpp:584:47: error: ‘class nvinfer1::INetworkDefinition’ has no member named ‘addConstant’
nvinfer1::IConstantLayer* preM = network->addConstant(preDims, pre);
^
trt_utils.cpp:611:5: error: ‘IConstantLayer’ is not a member of ‘nvinfer1’
nvinfer1::IConstantLayer* post_m = network->addConstant(postDims, post);
^
trt_utils.cpp:611:31: error: ‘post_m’ was not declared in this scope
nvinfer1::IConstantLayer* post_m = network->addConstant(postDims, post);
^
trt_utils.cpp:611:49: error: ‘class nvinfer1::INetworkDefinition’ has no member named ‘addConstant’
nvinfer1::IConstantLayer* post_m = network->addConstant(postDims, post);
^
trt_utils.cpp:616:5: error: ‘IMatrixMultiplyLayer’ is not a member of ‘nvinfer1’
nvinfer1::IMatrixMultiplyLayer* mm1
^
trt_utils.cpp:616:37: error: ‘mm1’ was not declared in this scope
nvinfer1::IMatrixMultiplyLayer* mm1
^
trt_utils.cpp:617:20: error: ‘class nvinfer1::INetworkDefinition’ has no member named ‘addMatrixMultiply’
= network->addMatrixMultiply(preM->getOutput(0), false, input, false);
^
trt_utils.cpp:621:5: error: ‘IMatrixMultiplyLayer’ is not a member of ‘nvinfer1’
nvinfer1::IMatrixMultiplyLayer mm2
^
trt_utils.cpp:621:37: error: ‘mm2’ was not declared in this scope
nvinfer1::IMatrixMultiplyLayer mm2
^
trt_utils.cpp:622:20: error: ‘class nvinfer1::INetworkDefinition’ has no member named ‘addMatrixMultiply’
= network->addMatrixMultiply(*mm1->getOutput(0), false, *post_m->getOutput(0), false);
^
Makefile:62: recipe for target ‘build/trt_utils.o’ failed
make: *** [build/trt_utils.o] Error 1

root@tx2:/usr/src/deepstream-plugins/sources/gst-yoloplugin/yoloplugin_lib# apt list --installed | grep -i tensor

WARNING: apt does not have a stable CLI interface. Use with caution in scripts.

nv-tensorrt-repo-ubuntu1604-ga-cuda9.0-trt3.0.4-20180208/now 1-1 arm64 [installed,local]
tensorrt/unknown,now 3.0.4-1+cuda9.0 arm64 [installed]
tensorrt-2.1.2/unknown,now 3.0.2-1+cuda8.0 arm64 [installed]

-albertr

albertr · September 27, 2018, 11:36pm

grep NV_TENSORRT /usr/include/aarch64-linux-gnu/NvInfer.h | head -3
#define NV_TENSORRT_MAJOR 3 //!< TensorRT major version
#define NV_TENSORRT_MINOR 0 //!< TensorRT minor version
#define NV_TENSORRT_PATCH 4 //!< TensorRT patch version

albertr · September 27, 2018, 11:38pm

Is TensorRT 4.x a requirement for this plugin?

-albertr

CJR · September 27, 2018, 11:43pm

@albertr Yes. Please install TensorRT 4 using jetpack 3.3.
Link to download Jetpack 3.3 - https://developer.nvidia.com/embedded/downloads#?search=jetpack%203.3

CJR · September 28, 2018, 12:51am

@prashanth.bhat The repo has now been updated with support for running yoloplugin on tegra with Deepstream 1.5

albertr · September 29, 2018, 4:45pm

Looks like this plugin wants source video in NV12 format?

root@tx2:/tmp/src/deepstream-plugins/sources/plugins/gst-yoloplugin-tegra# grep NV12 *
gstyoloplugin.cpp: GST_STATIC_CAPS(GST_VIDEO_CAPS_MAKE_WITH_FEATURES(GST_CAPS_FEATURE_MEMORY_NVMM, “{ NV12 }”)));
gstyoloplugin.cpp: GST_STATIC_CAPS(GST_VIDEO_CAPS_MAKE_WITH_FEATURES(GST_CAPS_FEATURE_MEMORY_NVMM, “{ NV12 }”)));

My USB camera supports I420, so it’s getting the following error:

------------> -----------------
** 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

Is there an easy way to add format conversion to gstreamer pipeline used by nvgstiva-app, something like adding nvvidconv element?

-albertr

Topic		Replies	Views
deepstream-plugins make failed DeepStream SDK	36	4769	September 7, 2022
Get wrong infer results while testing yolov4 on deepstream 5.0 DeepStream SDK	46	9384	October 12, 2021
Deepstream and JetPack 3.3 DeepStream SDK	33	5015	January 29, 2019
Python wrapper for tensorrt implementation of Yolo (currently v2) Jetson Nano	32	8015	July 2, 2020
Deep Learning Inference Benchmarking Instructions Jetson Nano	134	47563	May 30, 2023
DeepStream 4.0 + Custom Yolo Model - App run failed DeepStream SDK	10	4213	October 12, 2021
App Run Fails With Errors DeepStream SDK jetson-inference , yolo , debugging-and-troubleshooting	28	1770	October 23, 2023
Deepstrem 5.0 Python yolo DeepStream SDK	16	2671	October 12, 2021
Yolo for deepstream-app DeepStream SDK	27	9108	October 12, 2021
YoloV4 with OpenCV Jetson Nano yolo	12	6201	October 15, 2021

Yolo for Jetson

Related topics