openpose-1.3.0-win64-cpu-binaries is error：Cublas won't be available.

[nanako-ut]

I use openpose-1.3.0-win64-cpu-binaries.zip

when I run exe, it's error.

openpose-1.3.0-win64-cpu-binaries> bin/OpenPoseDemo.exe --video examples/media/video.avi
Starting pose estimation demo.
Starting thread(s)
E0423 15:19:01.718263 6652 common.cpp:123] Cannot create Cublas handle. Cublas won't be available.
E0423 15:19:01.729264 6652 common.cpp:130] Cannot create Curand generator. Curand won't be available.
F0423 15:19:01.737263 6652 cudnn_conv_layer.cpp:52] Check failed: error == cudaSuccess (35 vs. 0) CUDA driver version
is insufficient for CUDA runtime version
* Check failure stack trace: *
openpose-1.3.0-win64-cpu-binaries>

cpu-binaries needs cuda？

[nanako-ut]

1_extract_from_image.exe is same problem.

openposeopenpose-1.3.0-win64-cpu-binaries> bin/1_extract_from_image.exe
OpenPose Library Tutorial - Example 1.
E0425 14:45:45.005944 8760 common.cpp:123] Cannot create Cublas handle. Cublas won't be available.
E0425 14:45:45.021556 8760 common.cpp:130] Cannot create Curand generator. Curand won't be available.
F0425 14:45:45.037195 8760 cudnn_conv_layer.cpp:52] Check failed: error == cudaSuccess (35 vs. 0) CUDA driver version is insufficient for CUDA runtime version
* Check failure stack trace: *

[gineshidalgo99]

I'd try to debug the error, the problem is that my machine does have CUDA, so I can't detect this error on it.

As quick fix, could you try copying the CUDA DLLs from the GPU version for now? Does that work?

[gineshidalgo99]

Also, include your system configuration as it appears in the github issue template, it'd also help me a bit. THanks!

[nanako-ut]

I try copying 4 CUDA DLLs from openpose-1.3.0-win64-gpu-binaries.zip

target folder is openpose-1.3.0-win64-cpu-binariesbin
4 cuda dlls is
　・cublas64_80.dll
　・cudart64_80.dll
　・cudnn64_5.dll
　・curand64_80.dll

but the same error occured.
openpose-1.3.0-win64-cpu-binaries> bin/OpenPoseDemo.exe --video examples/media/video.avi
Starting pose estimation demo.
Starting thread(s)
E0429 21:13:00.251471 14404 common.cpp:123] Cannot create Cublas handle. Cublas won't be available.
E0429 21:13:00.268482 14404 common.cpp:130] Cannot create Curand generator. Curand won't be available.
F0429 21:13:00.292528 14404 cudnn_conv_layer.cpp:52] Check failed: error == cudaSuccess (35 vs. 0) CUDA driver version is insufficient for CUDA runtime version
* Check failure stack trace: *

Also, I did install caffe CPU version(Visual Studio 2015, CPU only, Python 3.5).

[wadereye]

I get the same error as you also.

[CharlesLiuyx]

I got the same error

Starting pose estimation demo.
Auto-detecting all available GPUs... Detected 1 GPU(s), using 1 of them starting at GPU 0.
Starting thread(s)
E0516 02:40:04.540171  9062 common.cpp:114] Cannot create Cublas handle. Cublas won't be available.
E0516 02:40:04.540477  9062 common.cpp:121] Cannot create Curand generator. Curand won't be available.
F0516 02:40:04.540578  9062 common.cpp:152] Check failed: error == cudaSuccess (35 vs. 0)  CUDA driver version is insufficient for CUDA runtime version
*** Check failure stack trace: ***
    @     0x7f85948f25cd  google::LogMessage::Fail()
    @     0x7f85948f4433  google::LogMessage::SendToLog()
    @     0x7f85948f215b  google::LogMessage::Flush()
    @     0x7f85948f4e1e  google::LogMessageFatal::~LogMessageFatal()
    @     0x7f8593e43cf2  caffe::Caffe::SetDevice()
    @     0x7f859618c30e  op::NetCaffe::initializationOnThread()
    @     0x7f85961b3bd8  op::addCaffeNetOnThread()
    @     0x7f85961b44f8  op::PoseExtractorCaffe::netInitializationOnThread()
    @     0x7f85961bc802  op::PoseExtractorNet::initializationOnThread()
    @     0x7f85961aff79  op::PoseExtractor::initializationOnThread()
    @     0x7f85961acd02  op::WPoseExtractor<>::initializationOnThread()
    @     0x7f85961e6b60  op::SubThread<>::initializationOnThread()
    @     0x7f85961e8d6f  op::Thread<>::initializationOnThread()
    @     0x7f85961e8fc1  op::Thread<>::threadFunction()
    @     0x7f8596202f47  std::_Mem_fn_base<>::operator()<>()
    @     0x7f8596202edb  _ZNSt12_Bind_simpleIFSt7_Mem_fnIMN2op6ThreadISt10shared_ptrISt6vectorINS1_5DatumESaIS5_EEES3_INS1_6WorkerIS8_EEEEEFvvEEPSC_EE9_M_invokeIJLm0EEEEvSt12_Index_tupleIJXspT_EEE
    @     0x7f8596202c8c  std::_Bind_simple<>::operator()()
    @     0x7f8596202a5c  std::thread::_Impl<>::_M_run()
    @     0x7f85957d7c80  (unknown)
    @     0x7f8594f296ba  start_thread
    @     0x7f859524641d  clone
    @              (nil)  (unknown)
[1]    9057 abort (core dumped)  ./build/examples/openpose/openpose.bin --image_dir examples/media/

I have cuda8 cuda9 cuda9.1 + cuDNN 7.1 6 5.1

And here is the version I got

> cat /usr/local/cuda/version.txt
CUDA Version 8.0.61

> cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2
#define CUDNN_MAJOR      5
#define CUDNN_MINOR      1
#define CUDNN_PATCHLEVEL 10
--
#define CUDNN_VERSION    (CUDNN_MAJOR * 1000 + CUDNN_MINOR * 100 + CUDNN_PATCHLEVEL)

#include "driver_types.h"



md5-bbb5ebec7e74eebc33d268b3a00ba6bd



Ubuntu 16.04 LTS
GTX 1080

[gjlroka]

I get the same error.I try copying 4 CUDA DLLs but the same error occured.
And according to https://github.com/CMU-Perceptual-Computing-Lab/openpose/issues/556 I shouldn't need to install CUDA and CUDNN for CPU.
I try to use it again without CUDA and CUDNN but the same error occured.

[gineshidalgo99]

@nanako-ut
@gjlroka
@repoalpha
@MaddalenaS
@CharlesLiuyx
@wadereye
@gkm2709

Sorry for this, I didn't have a computer without Nvidia graphic so I couldn't debug this earlier. I'll push a fix during this week. I'll keep you posted. Thanks

[jaskarannagi19]

Any progress here guys I am using openpose on cpu facing same errors

[MaddalenaS]

are there any news? @gineshidalgo99

[soulslicer]

it should be fixed this weekend

[gineshidalgo99]

Hi all,
@soulslicer
@nanako-ut
@gjlroka
@repoalpha
@MaddalenaS
@CharlesLiuyx
@wadereye
@gkm2709

Could you please try this provisional version? (done with the help of @soulslicer )
https://github.com/gineshidalgo99/openpose/tree/cpu

If it works, I will push it to OpenPose.

Again, we don't have a machine without Nvidia (CUDA) installed on it, so we can't detect all the errors, but we compiled a CPU-only version of Caffe, so it should work and not request any CUDA dependency. Please, let me know if it works, and if it doesn't, please post here the error. Thanks!

A proof for you to verify it was configured properly is that you should have a 3rdparty/windows/caffe_cpu folder after running CMake with CPU_ONLY option. Otherwise, let me know.

[jaskarannagi19]

Hi @gineshidalgo99

Thats what my cpu machine say to above. And I didnt get 3rdparty/windows/caffe_cpu this folder after running CMake-GUI on a windows 10 machine with VS 2015

D:openpose1>.buildx64debugOpenPoseDemo.exe --video examplesmediavideo.avi
Starting OpenPose demo...
Auto-detecting camera index... Detected and opened camera 0.
Starting thread(s)...
WARNING: Logging before InitGoogleLogging() is written to STDERR
E0528 06:49:23.461139 3832 common.cpp:123] Cannot create Cublas handle. Cublas won't be available.
E0528 06:49:23.462141 3832 common.cpp:130] Cannot create Curand generator. Curand won't be available.
I0528 06:49:23.467144 3832 upgrade_proto.cpp:67] Attempting to upgrade input file specified using deprecated input fields: modelspose/coco/pose_deploy_linevec.prototxt
I0528 06:49:23.467643 3832 upgrade_proto.cpp:70] Successfully upgraded file specified using deprecated input fields.
W0528 06:49:23.468144 3832 upgrade_proto.cpp:72] Note that future Caffe releases will only support input layers and not input fields.
I0528 06:49:23.472507 3832 net.cpp:51] Initializing net from parameters:
state {
phase: TEST
level: 0
}
layer {
name: "input"
type: "Input"
top: "image"
input_param {
shape {
dim: 1
dim: 3
dim: 1
dim: 1
}
}
}
..................................................................LOTS OF CON LAYERS IN BETWEEN...................................................................
convolution_param {
num_output: 128
pad: 3
kernel_size: 7
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"

I0528 06:49:23.484802 3832 layer_factory.cpp:58] Creating layer input
I0528 06:49:23.484802 3832 net.cpp:84] Creating Layer input
I0528 06:49:23.485301 3832 net.cpp:380] input -> image
I0528 06:49:23.485800 3832 net.cpp:122] Setting up input
I0528 06:49:23.486300 3832 net.cpp:129] Top shape: 1 3 1 1 (3)
I0528 06:49:23.486300 3832 net.cpp:137] Memory required for data: 12
I0528 06:49:23.486801 3832 layer_factory.cpp:58] Creating layer conv1_1
I0528 06:49:23.487301 3832 net.cpp:84] Creating Layer conv1_1
I0528 06:49:23.487301 3832 net.cpp:406] conv1_1 <- image
I0528 06:49:23.487802 3832 net.cpp:380] conv1_1 -> conv1_1
I0528 06:49:23.491305 3832 common.cpp:36] System entropy source not available, using fallback algorithm to generate seed instead.
F0528 06:49:23.491806 3832 cudnn_conv_layer.cpp:52] Check failed: error == cudaSuccess (35 vs. 0) CUDA driver version is insufficient for CUDA runtime version
* Check failure stack trace: *

[gineshidalgo99]

@jaskarannagi19 If you did not get 3rdparty/windows/caffe_cpu, that means you are not using the right repository, or that you didn't follow the CPU instructions (basically set CPU_ONLY in CMake).

Let me know your output after you can see that caffe_cpu is there. Thanks!

[jaskarannagi19]

@gineshidalgo99
Something like this happened...Yes Now I have that cpu folder

stacktrace.txt

[gineshidalgo99]

That's an out-of-memory (RAM) error (check FAQ and/or speed up hints). Please, reduce net_resolution to e.g. -1x80 to verify whether it runs properly other than the RAM memory issue.

[jaskarannagi19]

Hey @gineshidalgo99

Sorry for late reply but whenever I run this command it closes with not responding
.buildx64debugOpenPoseDemo.exe --video examplesmediavideo.avi --net_resolution -1x80
OR this
.buildx64debugOpenPoseDemo.exe --video examplesmediavideo.avi --net_resolution "320x320"
OR this
.buildx64debugOpenPoseDemo.exe --image_dir examplesmedia --net_resolution "320x320"

Then I decided to attach VS debugger than I received following error. By the way how do you debug this app Sorry for dump question I am totally new to all these things. This never runs on VS debugging mode

Unhandled exception at 0x00007FF968721630 (vcruntime140d.dll) in OpenPoseDemo.exe: 0xC0000005: Access violation reading location 0x0000000000000000.
If there is a handler for this exception, the program may be safely continued.

Even tried in cmd admin mode

[MaddalenaS]

@gineshidalgo99 I also tried the version you uploaded, but I was not able to run it.

First I tried to run it through Visual Studio after compiled it, but I obtained a long list of line regarding the layers and then program crashed. So, I was not able to read the errors.

Then I tried to run it from the cmd, but this time I obtained a series of missing .dll and so the program has not even started.

[gineshidalgo99]

Could you try the release version? In the debug one, caffe generates all that useless ouput for no reason. So it should only print the real error. Thanks! (Again, sorry but I don't have a non cuda machine to try it myself).

[MaddalenaS]

I had already tried the Release version before. But the problems are equal:
-compiling with visual studio the program crashed
-runnung in the cmd I had .dll problems.

Using CMake I put CPU_ONLY and I disabled USE_CUDNN.

[gineshidalgo99]

I am sorry for the error, and thanks for the feed-back. Then, we will need to wait until we can find a machine without CUDA on it. We can't do anything right now as we do not have this kind of hardware to replicate the issue. I will keep you posted.

The second solution is if you can compile Caffe for your Windows machine by yourselves, and then just indicate OpenPose where that Caffe is. The problem is Caffe not running for CPU_ONLY Windows machines.

[jaskarannagi19]

@gineshidalgo99
You can disable you card under device manager and booommm now you have a non cuda machine
OR
you can use my machine on share screen

[gineshidalgo99]

Our machines doesn't have integrated GPUs, making impossible to disable the Nvidia GPU (as I would lose my display). But we are looking for a machine without nvidia card, please, allow us 2-3 more days. Thanks!

[gineshidalgo99]

Hi, we have just finally tried it in a non-Nvidia machine, and everything worked fine with no errors.

Thus, the fix has been added to official OpenPose (download the latest OpenPose + follow CPU instructions).

If you are still having issues, they are not an OpenPose bug anymore, but rather an installation or Visual Studio issue. Some hints (also available in the doc):

1. Remove your OpenPose folder and re-download it completely.
2. Run net_resolution -1x80 to debug that the issue is not a RAM out of memory error.
3. Copy all DLLs from 3rdparty/caffe/.dll and 3rdparty/opencv/.dll to the same folder where OpenPoseDemo.exe is located.

If you are still having issues:

1. Post all your system hardware configuration (when you click in creating a new github issue you will see the template).
2. Add screenshots of the error itself.

Thanks for your debugging help!

[jaskarannagi19]

Thanks @gineshidalgo99
Its working from release build
No idea why it doesnt work in debug built.

[gineshidalgo99]

@jaskarannagi19 Thanks for the feed-back!

[whyky]

It works for me. Thank you.

[ssachio]

Do you have a plan to upload Modified binaries?

[gineshidalgo99]

With the next OpenPose version that should include many new interesting features. It should be released before June ends.

Binaries for 1.3.0 would never work, as that CPU version of OpenPose uses a Caffe that requires CUDA.

But you can always compile the latest code from GitHub yourself if you have (wanna download) Visual Studio.

[rnditdev]

Could you comment on the features to be released @gineshidalgo99? Thanks.

[MaddalenaS]

Sorry if I report another problem, but I'm the only one to get this error
(error.png) always regarding the OpenPose version for CPU only. I'm not
able to solve it. Maybe you can help me.
Thank you.

Maddalena

Il giorno gio 14 giu 2018 alle ore 20:16 Rodolfo Bonnin <
[email protected]> ha scritto:

Could you comment on the features to be released @gineshidalgo99
https://github.com/gineshidalgo99? Thanks.

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/CMU-Perceptual-Computing-Lab/openpose/issues/550#issuecomment-397390010,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AlkykJTUi-UdCwZW98GrQCDnlLTl1l47ks5t8qh-gaJpZM4Tfy02
.

[aarti9]

Hi @gineshidalgo99,
I am using the exe from the binaries but it gives me the error . Please advise. I need to run this urgently

C:openpose-1.3.0-win64-gpu-binaries>binOpenPoseDemo.exe --image_dir examplesmedia --face --hand --write_images exampleout
Starting pose estimation demo.

Error:
Cuda check failed (35 vs. 0): CUDA driver version is insufficient for CUDA runtime version

Coming from:

• D:UsersGinesDesktopopenpose-masteropenpose-1.3.0srcopenposegpucuda.cpp:op::getCudaGpuNumber():48
• D:UsersGinesDesktopopenpose-masteropenpose-1.3.0srcopenposegpucuda.cpp:op::cudaCheck():37
• D:UsersGinesDesktopopenpose-masteropenpose-1.3.0srcopenposegpucuda.cpp:op::getCudaGpuNumber():58
• D:UsersGinesDesktopopenpose-masteropenpose-1.3.0srcopenposegpugpu.cpp:op::getGpuNumber():27
• D:UsersGinesDesktopopenpose-masteropenpose-1.3.0includeopenpose/wrapper/wrapper.hpp:op::Wrapper >,class std::shared_ptr > > > >,class op::Queue > >,class std::queue > >,class std::deque > >,class std::allocator > > > > > > >::configure():941