Failed build TensorRT Yolo stand-alone app

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

I have Jetson TX2 with JetPack 3.3, TensorRT 4 and CUDA 9

dpkg -l | grep tensorrt
ii  tensorrt                                    4.0.2.0-1+cuda9.0                             arm64        Meta package of TensorRT

Following the tutorial here : https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps/tree/master/yolo , I got several problems.

First, the table on Readme file says that if I only want to use TRT stand alone app, I don’t need deepstream and CUDA 10.
jetson-TX1/TX2 trt-yolo-app Not required Jetpack 3.3 (Cuda 9.0, TensorRT 4, OpenCV 3.3)

But when I build using CMAKE, in CMakeLists.txt line 36 (https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps/blob/master/yolo/apps/trt-yolo/CMakeLists.txt) , you can see that CUDA 10 is required.

find_package(CUDA 10.0 EXACT REQUIRED cudart cublas curand)

Why is CUDA 10 required meanwhile the readme file says CUDA 9 is required?

Second, I changed the CMakeLists.txt so it will accept CUDA 9. When I did sudo make install it gave error like this :

In file included from /home/boulderai/deepstream_reference_apps/yolo/lib/ds_image.h:28:0,
                 from /home/boulderai/deepstream_reference_apps/yolo/lib/calibrator.h:29,
                 from /home/boulderai/deepstream_reference_apps/yolo/lib/calibrator.cpp:26:
/home/boulderai/deepstream_reference_apps/yolo/lib/trt_utils.h:84:22: error: ‘nvinfer1::DimsHW YoloTinyMaxpoolPaddingFormula::compute(nvinfer1::DimsHW, nvinfer1::DimsHW, nvinfer1::DimsHW, nvinfer1::DimsHW, nvinfer1::DimsHW, const char*) const’ marked ‘override’, but does not override
     nvinfer1::DimsHW compute(nvinfer1::DimsHW inputDims, nvinfer1::DimsHW kernelSize,
                      ^
compilation terminated due to -Wfatal-errors.
lib/CMakeFiles/yolo-lib.dir/build.make:69: recipe for target 'lib/CMakeFiles/yolo-lib.dir/calibrator.cpp.o' failed
make[2]: *** [lib/CMakeFiles/yolo-lib.dir/calibrator.cpp.o] Error 1
CMakeFiles/Makefile2:127: recipe for target 'lib/CMakeFiles/yolo-lib.dir/all' failed
make[1]: *** [lib/CMakeFiles/yolo-lib.dir/all] Error 2
Makefile:129: recipe for target 'all' failed
make: *** [all] Error 2

As far as I know, this problem is related to either Jetson Xavier or TensorRT 5, then why did I get this problem? Thank you.

2

I solved the problem by changing the repository to https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps/releases/tag/DS2.
I think it would be good if NVIDIA states to use the release version instead of master version cleary in the Readme file. I wrote the installation steps here https://gist.github.com/ivder/4d645935d65eb19bf27d99a8afeb33d8 or read the guide from the repo : https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps/tree/DS2

3

Hi,

You can find this information on the README file.

Platform	Applications	Deepstream Version	Dependancies	        Repo Release 
jetson-TX1/TX2	trt-yolo-app	Not required	        Jetpack 3.3 (...)	<b>DS2 Release</b>

Thanks.