some error on compiling

Question

some error on compiling

Opened this issue 8 years ago · 30 comments

Hello, faustomilletari

This is Jian Kang. I am very interested in your work 3D-caffe. So I tried to compile your 3D-Caffe, but I encountered a problem when I make runtest. it seems that something's wrong on softmax definition(dimension). Could you help me to fix it? The problem is as below. Thank you very much.

Best wishes,
Jian Kang

make -j all:
src/caffe/layers/softmax_loss_layer.cpp: In instantiation of ‘void caffe::SoftmaxWithLossLayer::Forward_cpu(const std::vectorcaffe::Blob<Dtype_>&, const std::vectorcaffe::Blob<Dtype_>&) [with Dtype = float]’:
src/caffe/layers/softmax_loss_layer.cpp:138:1: required from here
src/caffe/layers/softmax_loss_layer.cpp:60:16: warning: unused variable ‘weight’ [-Wunused-variable]
const Dtype* weight = bottom[2]->cpu_data();
src/caffe/layers/softmax_loss_layer.cpp: In instantiation of ‘void caffe::SoftmaxWithLossLayer::Forward_cpu(const std::vectorcaffe::Blob<Dtype_>&, const std::vectorcaffe::Blob<Dtype_>&) [with Dtype = double]’:
src/caffe/layers/softmax_loss_layer.cpp:138:1: required from here
src/caffe/layers/softmax_loss_layer.cpp:60:16: warning: unused variable ‘weight’ [-Wunused-variable]

make runtest:
22 tests from NetTest/0, where TypeParam = caffe::CPUDevice
[ RUN ] NetTest/0.TestGetLayerByName
F0912 05:07:01.832667 3518 layer.hpp:343] Check failed: ExactNumBottomBlobs() == bottom.size() (3 vs. 2) SoftmaxWithLoss Layer takes 3 bottom blob(s) as input.
*** Check failure stack trace: ***
@ 0x2aca2464ddaa (unknown)
@ 0x2aca2464dce4 (unknown)
@ 0x2aca2464d6e6 (unknown)
@ 0x2aca24650687 (unknown)
@ 0x461059 caffe::Layer<>::CheckBlobCounts()
@ 0x2aca25f71fd4 caffe::Net<>::Init()
@ 0x2aca25f73d62 caffe::Net<>::Net()
@ 0x575099 caffe::NetTest<>::InitNetFromProtoString()
@ 0x5328bc caffe::NetTest<>::InitTinyNet()
@ 0x56f4ba caffe::NetTest_TestGetLayerByName_Test<>::TestBody()
@ 0x890b23 testing::internal::HandleExceptionsInMethodIfSupported<>()
@ 0x887767 testing::Test::Run()
@ 0x88780e testing::TestInfo::Run()
@ 0x887915 testing::TestCase::Run()
@ 0x88ac58 testing::internal::UnitTestImpl::RunAllTests()
@ 0x88aee7 testing::UnitTest::Run()
@ 0x4514fa main
@ 0x2aca26e0af45 (unknown)
@ 0x457b59 (unknown)
@ (nil) (unknown)
make: *** [runtest] Aborted (core dumped)

Answer 1 · 2016-11-07T16:39:38.000Z

Same issue, any luck solving it?

Answer 2 · 2016-11-07T16:54:13.000Z

I think it was because you are not using cudnn.

Fausto Milletarì
Sent from my iPhone

On 7 Nov 2016, at 11:39, Luis Carlos Rivera notifications@github.com wrote:

Same issue, any luck solving it?

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub, or mute the thread.

Answer 3 · 2016-11-07T17:24:57.000Z

I am using cudnn but still get the same error, after running make all I get this issue

Answer 4 · 2016-11-14T14:09:33.000Z

I met the same issue, anyone solved it?

Answer 5 · 2016-11-14T15:31:46.000Z

could you please copy and paste here the full output and error message?

On Nov 14, 2016, at 9:09 AM, FayeYe notifications@github.com wrote:

I met the same issue, anyone solved it?

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub #1 (comment), or mute the thread https://github.com/notifications/unsubscribe-auth/AMtsvo-yFAsCQgwZYvq4966JLiiT8Qm6ks5q-GuegaJpZM4J7KPk.

Answer 6 · 2016-11-15T04:57:51.000Z

Which version of cuDNN are you using? there have been some method signature changes between v3 and v4.

Answer 7 · 2016-11-15T05:44:11.000Z

I use > v4

Fausto Milletarì
Sent from my iPhone

On 14 Nov 2016, at 23:57, Tisham Dhar notifications@github.com wrote:

Which version of cuDNN are you using? there have been some method signature changes between v3 and v4.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub, or mute the thread.

Answer 8 · 2016-11-15T19:15:22.000Z

This is the part of the make all output
src/caffe/layers/softmax_loss_layer.cpp: In instantiation of ‘void caffe::SoftmaxWithLossLayer<Dtype>::Forward_cpu(const std::vector<caffe::Blob<Dtype>*>&, const std::vector<caffe::Blob<Dtype>*>&) [with Dtype = float]’: src/caffe/layers/softmax_loss_layer.cpp:138:1: required from here src/caffe/layers/softmax_loss_layer.cpp:60:16: warning: unused variable ‘weight’ [-Wunused-variable] const Dtype* weight = bottom[2]->cpu_data(); ^ src/caffe/layers/softmax_loss_layer.cpp: In instantiation of ‘void caffe::SoftmaxWithLossLayer<Dtype>::Forward_cpu(const std::vector<caffe::Blob<Dtype>*>&, const std::vector<caffe::Blob<Dtype>*>&) [with Dtype = double]’: src/caffe/layers/softmax_loss_layer.cpp:138:1: required from here src/caffe/layers/softmax_loss_layer.cpp:60:16: warning: unused variable ‘weight’ [-Wunused-variable] CXX src/caffe/solver.cpp

Answer 9 · 2016-11-19T18:03:48.000Z

I can run it, verify that you had added the pythonpath of cudnn. I'm using cudnn v5

Answer 10 · 2016-11-21T06:32:25.000Z

Thanks for your reply! When I run make all ,it will output this warning. Ignoring the warning, I can run it.

Answer 11 · 2016-11-21T15:38:43.000Z

Please feel free to edit the code to get rid of the warning

Fausto Milletarì
Sent from my iPhone

On 21 Nov 2016, at 07:32, FayeYe notifications@github.com wrote:

Thanks for your reply! When I run make all ,it will output this warning. Ignoring the warning, I can run it.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub, or mute the thread.

Answer 12 · 2016-11-24T08:48:06.000Z

Met the same problem when "make runtest " by using cudnn 5.1 based on CUDA 8.0
Any comments ?

4 tests from SoftmaxWithLossLayerTest/2, where TypeParam = caffe::GPUDevice
[ RUN ] SoftmaxWithLossLayerTest/2.TestGradientIgnoreLabel
F1124 17:37:00.059399 18786 layer.hpp:343] Check failed: ExactNumBottomBlobs() == bottom.size() (3 vs. 2) SoftmaxWithLoss Layer takes 3 bottom blob(s) as input.
*** Check failure stack trace: ***
@ 0x7f7b8c32d5cd google::LogMessage::Fail()
@ 0x7f7b8c32f433 google::LogMessage::SendToLog()
@ 0x7f7b8c32d15b google::LogMessage::Flush()
@ 0x7f7b8c32fe1e google::LogMessageFatal::~LogMessageFatal()
@ 0xa75bcc caffe::GradientChecker<>::CheckGradientExhaustive()
@ 0xa7ad9f caffe::SoftmaxWithLossLayerTest_TestGradientIgnoreLabel_Test<>::TestBody()
@ 0xc45e03 testing::internal::HandleExceptionsInMethodIfSupported<>()
@ 0xc3eb9a testing::Test::Run()
@ 0xc3ece8 testing::TestInfo::Run()
@ 0xc3edf5 testing::TestCase::Run()
@ 0xc4098f testing::internal::UnitTestImpl::RunAllTests()
@ 0xc40cb3 testing::UnitTest::Run()
@ 0x7781a9 main
@ 0x7f7b83b79830 __libc_start_main
@ 0x77a2c9 _start

Answer 13 · 2016-11-24T09:13:18.000Z

The tests are not running for those layers because I changed the layer but I didn't update the tests. Fausto Milletarì Sent from my iPhone

…

On 24 Nov 2016, at 09:48, toki2015 ***@***.***> wrote: Met the same problem when "make runtest " by using cudnn 5.1 based on CUDA 8.0 Any comments ? 4 tests from SoftmaxWithLossLayerTest/2, where TypeParam = caffe::GPUDevice [ RUN ] SoftmaxWithLossLayerTest/2.TestGradientIgnoreLabel F1124 17:37:00.059399 18786 layer.hpp:343] Check failed: ExactNumBottomBlobs() == bottom.size() (3 vs. 2) SoftmaxWithLoss Layer takes 3 bottom blob(s) as input. *** Check failure stack trace: *** @ 0x7f7b8c32d5cd google::LogMessage::Fail() @ 0x7f7b8c32f433 google::LogMessage::SendToLog() @ 0x7f7b8c32d15b google::LogMessage::Flush() @ 0x7f7b8c32fe1e google::LogMessageFatal::~LogMessageFatal() @ 0xa75bcc caffe::GradientChecker<>::CheckGradientExhaustive() @ 0xa7ad9f caffe::SoftmaxWithLossLayerTest_TestGradientIgnoreLabel_Test<>::TestBody() @ 0xc45e03 testing::internal::HandleExceptionsInMethodIfSupported<>() @ 0xc3eb9a testing::Test::Run() @ 0xc3ece8 testing::TestInfo::Run() @ 0xc3edf5 testing::TestCase::Run() @ 0xc4098f testing::internal::UnitTestImpl::RunAllTests() @ 0xc40cb3 testing::UnitTest::Run() @ 0x7781a9 main @ 0x7f7b83b79830 __libc_start_main @ 0x77a2c9 _start — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

Answer 14 · 2016-11-24T09:19:54.000Z

You mean I can ignore this error ?
Thank you !

Answer 15 · 2016-11-24T09:23:24.000Z

You can Fausto Milletarì Sent from my iPhone

…

On 24 Nov 2016, at 10:19, toki2015 ***@***.***> wrote: You mean I can ignore this error ? Thank you ! — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

Answer 16 · 2016-12-16T23:01:55.000Z

Hi faustomilletari,

Thank you for the package! I have successfully installed 3D-caffe using "make all" with the warning you guys mentioned.

I also did "make pycaffe" and it is done too. However, when I try "import caffe" in Python, I can the following error message:
ImportError: /usr/lib/libboost_python.so.1.62.0: undefined symbol: PyUnicode_AsUTF8String

Any clue what is going on? Really appreciate!!!

Answer 17 · 2016-12-17T00:02:19.000Z

Seems something related with boost Fausto Milletarì Sent from my iPhone

…

On 16 Dec 2016, at 18:01, hzheng2015 ***@***.***> wrote: Hi faustomilletari, Thank you for the package! I have successfully installed 3D-caffe using "make all" with the warning you guys mentioned. I also did "make pycaffe" and it is done too. However, when I try "import caffe" in Python, I can the following error message: ImportError: /usr/lib/libboost_python.so.1.62.0: undefined symbol: PyUnicode_AsUTF8String Any clue what is going on? Really appreciate!!! — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

Answer 18 · 2016-12-17T04:46:09.000Z

Thanks Fausto! Could you let me know how did you install boost and libboost_python? what version?

Answer 19 · 2016-12-17T04:48:42.000Z

Here are the steps I followed to install 3D-caffe:

install all the dependencies
git clone https://github.com/faustomilletari/3D-Caffe
cp Makefile.config.example Makefile.config
make a few changes in Makefile.config
make all & make runtest
make pycaffe

Let me know if I am missing anything. Thanks!

Answer 20 · 2016-12-17T18:19:57.000Z

You can maybe refer to this tensorflow/tensorflow#479 the discussion refers to tensorflow but I think the findings apply to this case as well. Fausto Milletarì Sent from my iPhone

…

On 16 Dec 2016, at 23:48, hzheng2015 ***@***.***> wrote: Here are the steps I followed to install 3D-caffe: install all the dependencies git clone https://github.com/faustomilletari/3D-Caffe cp Makefile.config.example Makefile.config make a few changes in Makefile.config make all & make runtest make pycaffe Let me know if I am missing anything. Thanks! — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

Answer 21 · 2017-02-07T20:49:34.000Z

Hi

Instead of ATLAS if I use openBLAS ,will it be okay ? I know ATLAS is default but there is some issues with server.

thanks

Answer 22 · 2017-02-07T20:50:15.000Z

I guess there will be no difference. everything i’m concerned with runs using cudnn 5

…

On Feb 7, 2017, at 3:49 PM, sagarax009 ***@***.***> wrote: Hi Instead of ATLAS if I use openBLAS ,will it be okay ? I know ATLAS is default but there is some issues with server. thanks — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AMtsvufxRzVhg2E9zVxRL4W3JQhdx_Fqks5raNjegaJpZM4J7KPk>.

Answer 23 · 2017-02-07T20:54:19.000Z

Because there is system ATLAS ,and when I am compiling 3D vent it is giving error for ATLAS. I found solution of installlibatlas-base-dev but it downgrade the atlas version available on server.

what will be issue with cudnn 5 ,which I should use then?

Answer 24 · 2017-02-07T20:56:46.000Z

i think there will be no issue between atlas and cudnn5 so you are good to go

…

On Feb 7, 2017, at 3:54 PM, sagarax009 ***@***.***> wrote: Because there is system ATLAS ,and when I am compiling 3D vent it is giving error for ATLAS. I found solution of installlibatlas-base-dev but it downgrade the atlas version available on server. what will be issue with cudnn 5 ,which I should use then? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AMtsviGU9_UqNCSvIgAa7THgRhwlHJHpks5raNn8gaJpZM4J7KPk>.

Answer 25 · 2017-02-07T21:09:19.000Z

I am getting error
-latlas could not be found during make all. Is there any way to avoid ,without touching existing atlas?

Answer 26 · 2017-02-07T22:31:31.000Z

If cmake finds it then it has to work. You can refer to caffe documentation to install with Atlas. 3d caffe is not different at all from caffe in this aspect. Fausto Milletarì Sent from my iPhone

…

On 7 Feb 2017, at 16:09, sagarax009 ***@***.***> wrote: I am getting error -latlas could not be found during make all. Is there any way to avoid ,without touching existing atlas? — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

Answer 27 · 2017-02-07T22:43:33.000Z

Okay I will try ,or I will check with openBLAS ,hope it will work.Thanks

Answer 28 · 2017-02-08T23:22:31.000Z

Thanks man this build with cmake worked ...anyway I can train model without preprocessing right?

Or for pre processing which steps you normally used or do you have any script for .mhd files?

Answer 29 · 2017-02-09T01:15:30.000Z

You can try without pre processing:-) Fausto Milletarì Sent from my iPhone

…

On 8 Feb 2017, at 18:22, sagarax009 ***@***.***> wrote: Thanks man this build with cmake worked ...anyway I can train model without preprocessing right? Or for pre processing which steps you normally used or do you have any script for .mhd files? — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

Answer 30 · 2017-02-21T18:32:50.000Z

@JianKang1216 @toki2015
Hello I encountered the same issue when I run the "make runtest". This is how I solved it.

First the error looks like: Check failed: ExactNumBottomBlobs() == bottom.size() (3 vs. 2) SoftmaxWithLoss Layer takes 3 bottom blob(s) as input.

I think the error comes out cause the ExactNumBottomBlobs is set to be 3 in the loss_layer.hpp. However, for SoftmaxWithLoss layer the number of input blobs shall be 2: one is the label and the other one is the prediction blob. Therefore, simply change the ExactNumBottomBlobs to be 2 can solve this test error.

@faustomilletari Good job done for the 3D Caffe. Could you please edify this minor bug in the master version so that less people would be confused about this test error?

Thanks!
Zhuotun