Ncnn: 如何编译生成tensorflow2ncnn ?

[WenguoLi]

@nihui , 您好，请问如何编译生成tensorflow2ncnn？ 我看了所有相关的Issures，也按照其中的一些方法尝试了，但是还是没有成功编译，我不知道是不是需要有什么特殊的环境需要配置的？ 请问，可否提供一下的您的编译环境配置，让我做个参考。

多谢！

[sunbinbin1991]

+1

[XunChangqing]

tools/CMakeLists.txt
add_subdirectory(caffe)
add_subdirectory(mxnet)
Add
add_subdirectory(tensorflow)

[WenguoLi]

@XunChangqing ，多谢，我已经加了，可是build不通过，如下：

versions.proto:1:10: Unrecognized syntax identifier "proto3". This parser only recognizes "proto2".
tools/tensorflow/CMakeFiles/tensorflow2ncnn.dir/build.make:134: recipe for target 'tools/tensorflow/versions.pb.cc' failed
make[2]: * [tools/tensorflow/versions.pb.cc] Error 1
CMakeFiles/Makefile2:698: recipe for target 'tools/tensorflow/CMakeFiles/tensorflow2ncnn.dir/all' failed
make[1]: [tools/tensorflow/CMakeFiles/tensorflow2ncnn.dir/all] Error 2
Makefile:127: recipe for target 'all' failed
make: ** [all] Error 2

[WenguoLi]

@XunChangqing 您好，需要安装那个版本的protobuf，我安装了3.3的版本，还是不能build通过,.....
versions.pb.cc:(.text+0x137b): undefined reference to google::protobuf::io::CodedInputStream::ReadVarint32Fallback(unsigned int)' versions.pb.cc:(.text+0x13bf): undefined reference togoogle::protobuf::io::CodedInputStream::ReadVarint32Fallback(unsigned int)'
CMakeFiles/tensorflow2ncnn.dir/versions.pb.cc.o:(.data.rel.ro._ZTVN10tensorflow10VersionDefE[_ZTVN10tensorflow10VersionDefE]+0xb0): undefined reference to `google::protobuf::Message::SpaceUsedLong() const'
collect2: error: ld returned 1 exit status
tools/tensorflow/CMakeFiles/tensorflow2ncnn.dir/build.make:435: recipe for target 'tools/tensorflow/tensorflow2ncnn' failed
make[2]: * [tools/tensorflow/tensorflow2ncnn] Error 1
CMakeFiles/Makefile2:582: recipe for target 'tools/tensorflow/CMakeFiles/tensorflow2ncnn.dir/all' failed
make[1]: [tools/tensorflow/CMakeFiles/tensorflow2ncnn.dir/all] Error 2
Makefile:127: recipe for target 'all' failed
make: ** [all] Error 2

[WenguoLi]

问题已经解决了，给大家分享一下：

1. 需要安装3.0以上版本的protobuf，安装方法可以参考https://github.com/google/protobuf/blob/master/src/README.md
2. protobuf安装完成之后，protbuf的lib文件在/usr/lib下面，需要将文件拷贝到/usr/lib/x86_64-linux-gnu。
   cd /usr/lib
   sudo cp libproto* x86_64-linux-gnu/
3. 在tools/CMakeLists.txt中注释掉

add_subdirectory(caffe)

add_subdirectory(mxnet)

添加
add_subdirectory(tensorflow)
然后按照正常的编译流程编译即可！

不过，虽然编译过了，大多数的tensorflow模型还是无法转换，这个工具目前还是没有做完善。

[double-vane]

@WenguoLi 你好，我想问一下，你的可以用了吗？

[leeyevi]

我是3.4的protobuf也没编过
------ Build started: Project: caffe2ncnn, Configuration: Release x64 ------
caffe2ncnn.obj : error LNK2019: unresolved external symbol "private: static int google::protobuf::io::CodedInputStream::default_recursion_limit_" (?default_recursion_limit_@CodedInputStream@io@protobuf@google@@0HA) referenced in function "bool __cdecl read_proto_from_binary(char const *,class google::protobuf::Message *)" (?read_proto_from_binary@@YA_NPEBDPEAVMessage@protobuf@google@@@Z)
caffe.pb.obj : error LNK2019: unresolved external symbol "class google::protobuf::internal::ExplicitlyConstructed,class std::allocator > > google::protobuf::internal::fixed_address_empty_string" (?fixed_address_empty_string@internal@protobuf@google@@3V?$ExplicitlyConstructed@V?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@std@@@123@A) referenced in function "private: void __cdecl google::protobuf::internal::RepeatedPtrFieldBase::MergeFromInnerLoop::TypeHandler>(void * *,void * *,int,int)" (??$MergeFromInnerLoop@VTypeHandler@?$RepeatedPtrField@VParamSpec@caffe@@@protobuf@google@@@RepeatedPtrFieldBase@internal@protobuf@google@@AEAAXPEAPEAX0HH@Z)
D:projectncnnbuildtoolscaffeReleasecaffe2ncnn.exe : fatal error LNK1120: 2 unresolved externals
------ Build started: Project: onnx2ncnn, Configuration: Release x64 ------
onnx2ncnn.obj : error LNK2019: unresolved external symbol "private: static int google::protobuf::io::CodedInputStream::default_recursion_limit_" (?default_recursion_limit_@CodedInputStream@io@protobuf@google@@0HA) referenced in function "bool __cdecl read_proto_from_binary(char const *,class google::protobuf::Message *)" (?read_proto_from_binary@@YA_NPEBDPEAVMessage@protobuf@google@@@Z)
onnx.pb.obj : error LNK2019: unresolved external symbol "class google::protobuf::internal::ExplicitlyConstructed,class std::allocator > > google::protobuf::internal::fixed_address_empty_string" (?fixed_address_empty_string@internal@protobuf@google@@3V?$ExplicitlyConstructed@V?$basic_string@DU?$char_traits@D@std@@V?$allocator@D@2@@std@@@123@A) referenced in function "protected: void __cdecl google::protobuf::internal::RepeatedPtrFieldBase::Clear::TypeHandler>(void)" (??$Clear@VTypeHandler@?$RepeatedPtrField@VTensorShapeProto_Dimension@onnx@@@protobuf@google@@@RepeatedPtrFieldBase@internal@protobuf@google@@IEAAXXZ)
D:projectncnnbuildtoolsonnxReleaseonnx2ncnn.exe : fatal error LNK1120: 2 unresolved externals
========== Build: 0 succeeded, 2 failed, 5 up-to-date, 0 skipped ==========

[HudsonHuang]

I found an nice blog about this, see:
https://blog.csdn.net/u012477435/article/details/80765005

[XHuaSheng]

我用的是3.6.1版本的protobuf 不过用的是静态库，可以一下编过，不过没搞明白咋用。。。

[nihui]

duplicate of https://github.com/Tencent/ncnn/issues/5