把mxnet模型转onnx模型前需要Note out（注释掉） the line 57(predict_score = mxnet.symbol.slice_axis(predict_score, axis=1, begin=0, end=1)，现在我想用python+tensorrt推理onnx模型，如何得到注释the line 57之前的结果呢？

to run test.cppbut and error output below:

Create interpreter failed, open symbol_10_560_25L_8scales_v1_deploy.mnn/symbol_10_320_20L_5scales_v2_deploy.mnn error

using param:

• symbol_10_560_25L_8scales_v1_deploy.mnn
• test_5.jpg

PLATFORM

• ubuntu 16.04 , x86
• linking libMNN.so
• QT

hi, have you plan?

Hello, I probably don't know much about c++. When I compiled test.cpp incorrectly, my command was g++ test.cpp-std=c++ 11 pkg-config -- LIBS -- cflags opencv4-I. /src. I link MNN libraries and header files at first level

在mac上编译运行成功了，想测试一下使用opencl是否会更快一点，但是编译报错
Undefined symbols for architecture x86_64:
"MNN::Interpreter::releaseModel()", referenced from:
LFFD::~LFFD() in MNN_LFFD.cpp.o
"MNN::Interpreter::resizeTensor(MNN::Tensor*, std::__1::vector<int, std::__1::allocator > const&)", referenced from:
LFFD::detect(cv::Mat&, std::__1::vector<FaceInfo, std::__1::allocator >&, int, int, float, float, int, std::__1::vector<int, std::__1::allocator >) in MNN_LFFD.cpp.o
"MNN::Interpreter::createSession(MNN::ScheduleConfig const&)", referenced from:
LFFD::LFFD(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&, int, int) in MNN_LFFD.cpp.o
"MNN::Interpreter::resizeSession(MNN::Session*)", referenced from:
LFFD::detect(cv::Mat&, std::__1::vector<FaceInfo, std::__1::allocator >&, int, int, float, float, int, std::__1::vector<int, std::__1::allocator >) in MNN_LFFD.cpp.o
"MNN::Interpreter::createFromFile(char const*)", referenced from:
LFFD::LFFD(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&, int, int) in MNN_LFFD.cpp.o
"MNN::Interpreter::releaseSession(MNN::Session*)", referenced from:
LFFD::~LFFD() in MNN_LFFD.cpp.o
"MNN::Interpreter::getSessionInput(MNN::Session const*, char const*)", referenced from:
LFFD::LFFD(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&, int, int) in MNN_LFFD.cpp.o
"MNN::Interpreter::getSessionOutput(MNN::Session const*, char const*)", referenced from:
LFFD::LFFD(std::__1::basic_string<char, std::__1::char_traits, std::__1::allocator > const&, int, int) in MNN_LFFD.cpp.o
"MNN::Interpreter::~Interpreter()", referenced from:
std::__1::default_deleteMNN::Interpreter::operator()(MNN::Interpreter*) const in MNN_LFFD.cpp.o
"MNN::CV::ImageProcess::create(MNN::CV::ImageProcess::Config const&, MNN::Tensor const*)", referenced from:
LFFD::detect(cv::Mat&, std::__1::vector<FaceInfo, std::__1::allocator >&, int, int, float, float, int, std::__1::vector<int, std::__1::allocator >) in MNN_LFFD.cpp.o
"MNN::CV::ImageProcess::convert(unsigned char const*, int, int, int, MNN::Tensor*)", referenced from:
LFFD::detect(cv::Mat&, std::__1::vector<FaceInfo, std::__1::allocator >&, int, int, float, float, int, std::__1::vector<int, std::__1::allocator >) in MNN_LFFD.cpp.o
"MNN::CV::ImageProcess::~ImageProcess()", referenced from:
std::__1::default_deleteMNN::CV::ImageProcess::operator()(MNN::CV::ImageProcess*) const in MNN_LFFD.cpp.o
"MNN::Tensor::Tensor(MNN::Tensor const*, MNN::Tensor::DimensionType, bool)", referenced from:
LFFD::detect(cv::Mat&, std::__1::vector<FaceInfo, std::__1::allocator >&, int, int, float, float, int, std::__1::vector<int, std::__1::allocator >) in MNN_LFFD.cpp.o
"MNN::Tensor::~Tensor()", referenced from:
LFFD::detect(cv::Mat&, std::__1::vector<FaceInfo, std::__1::allocator >&, int, int, float, float, int, std::__1::vector<int, std::__1::allocator >) in MNN_LFFD.cpp.o
"MNN::Interpreter::runSession(MNN::Session*) const", referenced from:
LFFD::detect(cv::Mat&, std::__1::vector<FaceInfo, std::__1::allocator >&, int, int, float, float, int, std::__1::vector<int, std::__1::allocator >) in MNN_LFFD.cpp.o
"MNN::Tensor::copyToHostTensor(MNN::Tensor*) const", referenced from:
LFFD::detect(cv::Mat&, std::__1::vector<FaceInfo, std::__1::allocator >&, int, int, float, float, int, std::__1::vector<int, std::__1::allocator >) in MNN_LFFD.cpp.o
"MNN::Tensor::getDimensionType() const", referenced from:
MNN::Tensor::width() const in MNN_LFFD.cpp.o
MNN::Tensor::height() const in MNN_LFFD.cpp.o
"MNN::Tensor::shape() const", referenced from:
LFFD::detect(cv::Mat&, std::__1::vector<FaceInfo, std::__1::allocator >&, int, int, float, float, int, std::__1::vector<int, std::__1::allocator >) in MNN_LFFD.cpp.o
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
make[2]: *** [test] Error 1
make[1]: *** [CMakeFiles/test.dir/all] Error 2
make: *** [all] Error 2
请问你做了OpenCL或者vulkan后端的测试了吗，能否解答一下疑惑，谢谢

I know is not this repo issue, but how did u guys able to convert to onnx? with original repo, I got some error:

INFO:root:Output node is: conv20_3_bbox
Traceback (most recent call last):
  File "to_onnx.py", line 39, in <module>
    generate_onnx_file()
  File "to_onnx.py", line 29, in generate_onnx_file
    onnx_mxnet.export_model(net_symbol, net_params, [input_shape], numpy.float32, onnx_path, verbose=True)
  File "/usr/local/lib/python3.5/dist-packages/mxnet/contrib/onnx/mx2onnx/export_model.py", line 87, in export_model
    verbose=verbose)
  File "/usr/local/lib/python3.5/dist-packages/mxnet/contrib/onnx/mx2onnx/export_onnx.py", line 312, in create_onnx_graph_proto
    checker.check_graph(graph)
  File "/usr/local/lib/python3.5/dist-packages/onnx/checker.py", line 53, in checker
    proto.SerializeToString(), ctx)
onnx.onnx_cpp2py_export.checker.ValidationError: Node (slice_axis20) has input size 1 not in range [min=3, max=5].

==> Context: Bad node spec: input: "softmax0" output: "slice_axis20" name: "slice_axis20" op_type: "Slice" attribute { name: "axes" ints: 1 type: INTS } attribute { name: "ends" ints: 1 type: INTS } attribute { name: "starts" ints: 0 type: INTS }

I am using the every newest version of mxnet and onnx

您好，下载您的源码后，Ubuntu16.04上成功编译了，但进行测试时就报了Segmentation fault (core

• dumped)的错误，debug后发现在sess_lffd = lffd->createSession(config);就莫名地报错停止了，请教您如何解决这个问题呢？

Thank you for your sharing~ Can you give time consumption statistics on pc or mobile?

Thank you very much for your efforts and contributions.

您好，您的源码可以在树莓派上正常编译和运行，不过，推理的时间较长，需要2s多；
如果我想在windows上测试，该如何修改？我修改了cmakelists文件如下：能够编译，但是运行的时候，报错：缺少MNN.dll和opencv_world410d.dll，请教正确的cmakelists文件该如何修改？
另外，我看到您的博客，讲如何在windows平台上编译转换工具，非常详细，但是我按照步骤进行下来报错：
一开始报运行时版本不匹配，我在vs里面修改了运行时以后，编译到99%的时候链接失败。。。
#1.cmake verson
cmake_minimum_required(VERSION 2.8)

#2.project name
project(test C CXX)

#opencv
set(OpenCV_DIR C:/Users/YuKai/Downloads/opencv/opencv/build)
#set(OpenCV_DIR env{OpenCV_DIR})
find_package(OpenCV REQUIRED)

#3.set environment variable
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")

#4.mnn include
include_directories(${CMAKE_CURRENT_LIST_DIR}/MNN/include)

mnn lib

add_library( MNN SHARED IMPORTED )
set_target_properties(
MNN
PROPERTIES IMPORTED_IMPLIB
${CMAKE_CURRENT_LIST_DIR}/MNN/lib/MNN.lib
)

#6.source directory
file(GLOB TEST_SRC ${CMAKE_CURRENT_LIST_DIR}/src/.h
${CMAKE_CURRENT_LIST_DIR}/src/.cpp)
set(TEST_COMPILE_CODE ${TEST_SRC})

add_executable(test ${TEST_COMPILE_CODE})

target_link_libraries(test
MNN
${OpenCV_LIBS}
)

你好
我参考face detection 的方法转了head detection，得到了onnx和mnn模型

mnn模型检测的时候，没有框输出

dyld: Library not loaded: /usr/local/opt/openssl/lib/libssl.1.0.0.dylib
Referenced from: /usr/local/opt/ffmpeg/lib/libavcodec.57.dylib
Reason: image not found
[1] 19815 abort ./test ../models ../data/demo.jpg
请问这是什么问题