These are chat archives for beniz/deepdetect

14th
Feb 2017
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 02:45
@ArdalanM , I got the same error while compiling with Tensorflow even after the LIBRAY_PATH
blob
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 02:53
blob
Emmanuel Benazera
@beniz
Feb 14 2017 08:36
@sunatthegilddotcom you may want to post the full error if you have not solved it.
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 08:47

@beniz . Thank you very much. I have not solved it yet. Your help is greatl y appreciated. Followed the exact steps on your github site. The only variable could be Bazel version. I had issue with Bazel 0.4.4, I roll back to 0.4.3. Here are the error messages.
/home/wsun/deepdetect/build/protobuf/include/google/protobuf/map.h:1013:9: required from ‘void google::protobuf::Map<Key, T>::InnerMap::erase(google::protobuf::Map<Key, T>::InnerMap::iterator) [with Key = std::basic_string<char>; T = int; google::protobuf::Map<Key, T>::InnerMap::iterator = google::protobuf::Map<std::basic_string<char>, int>::InnerMap::iterator_base<google::protobuf::Map<std::basic_string<char>, int>::KeyValuePair>]’
/home/wsun/deepdetect/build/protobuf/include/google/protobuf/map.h:1612:7: required from ‘google::protobuf::Map<Key, T>::iterator google::protobuf::Map<Key, T>::erase(google::protobuf::Map<Key, T>::iterator) [with Key = std::basic_string<char>; T = int]’
/home/wsun/deepdetect/build/protobuf/include/google/protobuf/map.h:1617:26: required from ‘void google::protobuf::Map<Key, T>::erase(google::protobuf::Map<Key, T>::iterator, google::protobuf::Map<Key, T>::iterator) [with Key = std::basic_string<char>; T = int]’
/home/wsun/deepdetect/build/protobuf/include/google/protobuf/map.h:1620:38: required from ‘void google::protobuf::Map<Key, T>::clear() [with Key = std::basic_string<char>; T = int]’
/home/wsun/deepdetect/build/protobuf/include/google/protobuf/map_field_inl.h:228:3: required from ‘void google::protobuf::internal::MapField<Key, T, key_wire_type, value_wire_type, default_enum_value>::Clear() [with Key = std::basic_string<char>; T = int; google::protobuf::internal::WireFormatLite::FieldType kKeyFieldType = (google::protobuf::internal::WireFormatLite::FieldType)9u; google::protobuf::internal::WireFormatLite::FieldType kValueFieldType = (google::protobuf::internal::WireFormatLite::FieldType)5u; int default_enum_value = 0]’
/home/wsun/deepdetect/build/tensorflow/src/tensorflow/dbuild/execroot/tensorflow/bazel-out/host/genfiles/tensorflow/core/protobuf/config.pb.h:1763:23: required from here
/home/wsun/deepdetect/build/protobuf/include/google/protobuf/map.h:606:10: warning: unused parameter ‘n’ [-Wunused-parameter]
make2: [src/CMakeFiles/ddetect.dir/deepdetect.cc.o] Error 1
make1:
[src/CMakeFiles/ddetect.dir/all] Error 2
make: * [all] Error 2
wsun@deeplearningdetect03:~/deepdetect/build$

How can I get more detail on the error? Thanks a lot!

Emmanuel Benazera
@beniz
Feb 14 2017 08:48
yes bazel 4.4 is rotten with TF, AFAIK, bazel 4.3 is OK
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 08:48
@beniz , thanks a lot.
Emmanuel Benazera
@beniz
Feb 14 2017 08:49
your log above does not contain the error, it must be higher up. Put it all into a gist maybe and post the link here.
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 08:54
@beniz ok. Let me get all of the log. Pull it all into the gist. Thanks
Emmanuel Benazera
@beniz
Feb 14 2017 09:14
are you on Azure ?
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 09:48
@beniz , Yes, I am in Azure
@beniz Thank you very much for your help
Emmanuel Benazera
@beniz
Feb 14 2017 09:55
so, this is because you've built Caffe by going into the caffe_dd repository and typing make. You actually cannot do this while building with TF support as both TF and Caffe need to share the exact same version of protobuf
Instead, you should modify deepdetect/CMakeLists.txt and remove the -j${N} from the caffe build.
regarding Azure, are you using the new series N VM with GPU ? I couldn't figure out the exact pricing and path in order to provide pre-built VMs on Azure. We are providing pre-built AMIs on Amazon AWS with built-in models, TF+Xgboost+Caffe and I'm interested to understand whether Azure is cheaper, and/or better in any ways.
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 10:01
@beniz , I have not used GPU yet. But I will figure that out and report back.
Emmanuel Benazera
@beniz
Feb 14 2017 10:03
@sunatthegilddotcom then if you are running on CPU, let me know whether this is cheaper than AWS or better in your opinion :)
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 10:09
@beniz , I have not compared the price yet. From my test, Azure server's performance is more uniformed vs. AWS EC2.
@beniz, Please forgive my ignorance, Is this the right place to remove the -j${N}
  else() # USE_TF
    ExternalProject_Add(
      caffe_dd
      PREFIX caffe_dd
      INSTALL_DIR ${CMAKE_BINARY_DIR}
      URL https://github.com/beniz/caffe/archive/master.tar.gz
      CONFIGURE_COMMAND ln -sf Makefile.config.gpu.cudnn Makefile.config && echo "OPENCV_VERSION:=${OPENCV_VERSION}" >> Makefile.config && make -j${N}
      INSTALL_COMMAND ""
      BUILD_IN_SOURCE 1
      )
  endif()
else() # HAVE_CUDNN
Emmanuel Benazera
@beniz
Feb 14 2017 10:11
remove them everywhere, that'll be safer
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 10:12
@beniz , thank you very much. Will do
Emmanuel Benazera
@beniz
Feb 14 2017 10:12
how long did your TF build take ? the problem above suggests your machine is a bit slow, or busy.
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 10:12
@beniz, I will also research the price for GPU and GPU servers and see whether Azure is cheaper
Emmanuel Benazera
@beniz
Feb 14 2017 10:14
look, I don't think I've ever advertized much on this channel, but when I see people using external clouds and struggling with the builds, I believe it is relevant to point to the pre-build AMIs, the CPU one is here: https://aws.amazon.com/marketplace/pp/B01N1RGWQZ
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 10:14
@beniz, it is fairly quick. It took a few minutes to get to the error
Emmanuel Benazera
@beniz
Feb 14 2017 10:14
We've rolled a new version out yesterdays, that includes object detection.
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 10:15
@beniz that's great. I think you should do one for Azure as well.
Emmanuel Benazera
@beniz
Feb 14 2017 10:15
@sunatthegilddotcom few minutes means you are not building TF. TF build takes ~30mins to 45mins on an average server.
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 10:18
@beniz, I have been struggling to build TF for a while, I got this error, I was trying to figure it out myself before posting. It is half-built already, It is not from very begining
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 12:54
@beniz , I got the same error. Here is the whole log https://gist.github.com/sunatthegilddotcom/4195cfb653ed323ac6d41c0370af7792
Emmanuel Benazera
@beniz
Feb 14 2017 13:01
have you started from a fresh build ? what is your exact cmake call ? Also, please state your exact VM OS (Ubuntu 14.04 or 16.04).
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 13:02
@beniz , I think it might be better from a Fresh build. it is Ubuntu 14.04
Emmanuel Benazera
@beniz
Feb 14 2017 13:18
yes, fresh build is best. TF will build, then Caffe. In this case the Caffe build uses the protobuf from TF. Remove the -j${N}.
sunatthegilddotcom
@sunatthegilddotcom
Feb 14 2017 13:52
@beniz , Freshcbuild is passed flawless
@beniz . Thank you very much!
Emmanuel Benazera
@beniz
Feb 14 2017 13:52
good.