YOLOv9 build error

[reaganch]

Env

• GPU, e.g. V100, RTX2080, TX2, Xavier NX, Nano, etc.
  GTX 1070

• OS, e.g. Ubuntu16.04, Win10, etc.
  Linux Mint 21.3

• Cuda version
  Cuda version 12.3

• TensorRT version
  TensorRT version 10.0

About this repo

• which branch/tag/commit are you using?
  Latest

• which model? yolov5, retinaface?
  yolov9

Your problem

• what is your command? e.g. sudo ./yolov5 -s
  I'm running make

• what's your output?
  I get the following error:

% make
[ 20%] Built target myplugins
[ 30%] Building CXX object CMakeFiles/yolov9.dir/demo.cpp.o
In file included from /usr/local/include/opencv4/opencv2/core/vsx_utils.hpp:11,
                 from /usr/local/include/opencv4/opencv2/core/base.hpp:661,
                 from /usr/local/include/opencv4/opencv2/core.hpp:53,
                 from /usr/local/include/opencv4/opencv2/opencv.hpp:52,
                 from /home/cricket/build/tensorrtx/yolov9/include/postprocess.h:4,
                 from /home/cricket/build/tensorrtx/yolov9/demo.cpp:7:
/home/cricket/build/tensorrtx/yolov9/demo.cpp: In function ‘void prepare_buffer(nvinfer1::ICudaEngine*, float**, float**, float**)’:
/home/cricket/build/tensorrtx/yolov9/demo.cpp:72:20: error: ‘class nvinfer1::ICudaEngine’ has no member named ‘getNbBindings’
   72 |     assert(engine->getNbBindings() == 2);
      |                    ^~~~~~~~~~~~~
/home/cricket/build/tensorrtx/yolov9/demo.cpp:75:36: error: ‘class nvinfer1::ICudaEngine’ has no member named ‘getBindingIndex’
   75 |     const int inputIndex = engine->getBindingIndex(kInputTensorName);
      |                                    ^~~~~~~~~~~~~~~
/home/cricket/build/tensorrtx/yolov9/demo.cpp:76:37: error: ‘class nvinfer1::ICudaEngine’ has no member named ‘getBindingIndex’
   76 |     const int outputIndex = engine->getBindingIndex(kOutputTensorName);
      |                                     ^~~~~~~~~~~~~~~
/home/cricket/build/tensorrtx/yolov9/demo.cpp: In function ‘void infer(nvinfer1::IExecutionContext&, CUstream_st*&, void**, float*, int)’:
/home/cricket/build/tensorrtx/yolov9/demo.cpp:88:13: error: ‘class nvinfer1::IExecutionContext’ has no member named ‘enqueue’; did you mean ‘enqueueV3’?
   88 |     context.enqueue(batchSize, buffers, stream, nullptr);
      |             ^~~~~~~
      |             enqueueV3
make[2]: *** [CMakeFiles/yolov9.dir/build.make:76: CMakeFiles/yolov9.dir/demo.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:111: CMakeFiles/yolov9.dir/all] Error 2
make: *** [Makefile:91: all] Error 2

• what output do you expect?
  I expect the build to complete without any errors.

[reaganch]

I presume this may be because TensorRTX needs an older version of TensorRT installed? Would be great if you could confirm. Thanks!

[wang-xinyu]

Yes, try to use TensorRT <= 8.5

[reaganch]

Thanks for that. Will give it a shot. Cheers!

[reaganch]

Just installed TensorRT version 8.5 GA Update 2. This required installing CUDA version 11.8.0 and cuDNN version 8.9.7. I had previously installed CUDA version 12.3.2 to build OpenCV, so I currently have two installations of CUDA on my system. When I try to build TensorRTX for yolov9, I now get the following error when I run cmake ... Could you please advise what I may be doing wrong here and why it seems to be requiring CUDA 12.3?

% cmake ..
-- The C compiler identification is GNU 11.4.0
-- The CXX compiler identification is GNU 11.4.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /usr/bin/cc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/c++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- operation system is Linux-6.5.0-26-generic
-- current platform: Linux 
-- The CUDA compiler identification is NVIDIA 11.8.89
-- Detecting CUDA compiler ABI info
-- Detecting CUDA compiler ABI info - done
-- Check for working CUDA compiler: /usr/local/cuda/bin/nvcc - skipped
-- Detecting CUDA compile features
-- Detecting CUDA compile features - done
CMake Warning (dev) at /usr/local/lib/cmake/opencv4/OpenCVConfig.cmake:86 (find_package):
  Policy CMP0146 is not set: The FindCUDA module is removed.  Run "cmake
  --help-policy CMP0146" for policy details.  Use the cmake_policy command to
  set the policy and suppress this warning.

Call Stack (most recent call first):
  /usr/local/lib/cmake/opencv4/OpenCVConfig.cmake:108 (find_host_package)
  CMakeLists.txt:41 (find_package)
This warning is for project developers.  Use -Wno-dev to suppress it.

-- Performing Test CMAKE_HAVE_LIBC_PTHREAD
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Success
-- Found Threads: TRUE
CMake Error at /home/cricket/.local/lib/python3.10/site-packages/cmake/data/share/cmake-3.29/Modules/FindPackageHandleStandardArgs.cmake:230 (message):
  Could NOT find CUDA: Found unsuitable version "11.8", but required is exact
  version "12.3" (found /usr/local/cuda-11.8)
Call Stack (most recent call first):
  /home/cricket/.local/lib/python3.10/site-packages/cmake/data/share/cmake-3.29/Modules/FindPackageHandleStandardArgs.cmake:598 (_FPHSA_FAILURE_MESSAGE)
  /home/cricket/.local/lib/python3.10/site-packages/cmake/data/share/cmake-3.29/Modules/FindCUDA.cmake:1291 (find_package_handle_standard_args)
  /usr/local/lib/cmake/opencv4/OpenCVConfig.cmake:86 (find_package)
  /usr/local/lib/cmake/opencv4/OpenCVConfig.cmake:108 (find_host_package)
  CMakeLists.txt:41 (find_package)


-- Configuring incomplete, errors occurred!

Thanks!

[wang-xinyu]

Seems your opencv is linking cuda12.3, you can try to use docker

[reaganch]

Ah, I see. Thanks for the suggestion. Will give that a shot.

[zmtttt]

root@b37d8b2aacd1:/workspace/tensorrtx/yolov9/build# sudo ./yolov9 -s ../yolov9-c.wts yolov9-c.engine c
[04/22/2024-13:15:57] [W] [TRT] The implicit batch dimension mode has been deprecated. Please create the network with NetworkDefinitionCreationFlag::kEXPLICIT_BATCH flag whenever possible.
Loading weights: ../yolov9-c.wts
Your platform support int8: true
Building engine, please wait for a while...
reading calib cache: int8calib.table
[04/22/2024-13:16:02] [W] [TRT] TensorRT was linked against cuDNN 8.6.0 but loaded cuDNN 8.5.0
[04/22/2024-13:16:03] [W] [TRT] TensorRT was linked against cuDNN 8.6.0 but loaded cuDNN 8.5.0
[04/22/2024-13:16:03] [W] [TRT] TensorRT was linked against cuDNN 8.6.0 but loaded cuDNN 8.5.0
[04/22/2024-13:16:03] [E] [TRT] 1: Unexpected exception _Map_base::at
[04/22/2024-13:16:03] [E] [TRT] 2: [builder.cpp::buildSerializedNetwork::751] Error Code 2: Internal Error (Assertion engine != nullptr failed. )
Build engine successfully!
yolov9: /workspace/tensorrtx/yolov9/demo.cpp:31: void serialize_engine(unsigned int, std::string&, std::string&, std::string&): Assertion `serialized_engine != nullptr' failed.
Aborted

[wang-xinyu]

@zmtttt have you check this path

tensorrtx/yolov9/include/config.h

Line 13 in d4aa52d

const static char* gCalibTablePath = "./calib";

[wang-xinyu]

@WuxinrongY Can we make yolov9 to use fp16 by default?

[WuxinrongY]

@WuxinrongY Can we make yolov9 to use fp16 by default?

好

[zmtttt]

@wang-xinyu ，谢谢，不过const static char* gCalibTablePath = "/home/zhaomt/com/tensorrtx/yolov9/calib/coco_calib"，，还是同样的错误

[WuxinrongY]

@wang-xinyu ，谢谢，不过const static char* gCalibTablePath = "/home/zhaomt/com/tensorrtx/yolov9/calib/coco_calib"，，还是同样的错误

这个路径后面要加“/”，比如const static char* gCalibTablePath = "/home/zhaomt/com/tensorrtx/yolov9/calib/coco_calib/"