search

davisking / dlib

A toolkit for making real world machine learning and data analysis applications in C++
http://dlib.net
Boost Software License 1.0
13.57k stars 3.38k forks source link

[Bug]: python setup.py install (_dlib_pybind11: not found error) / pip install dlib (CUDA disabled error) #2979

Open eddiehe99 opened 4 months ago

eddiehe99 commented 4 months ago

What Operating System(s) are you seeing this problem on?

Windows

dlib version

19.24.0 19.24.1 19.24.2 19.24.3 19.24.4

Python version

3.7 3.8 3.9 3.10 3.11 3.12

Compiler

MSVC 19

Expected Behavior

import dlib without errors dlib.DLIB_USE_CUDA returns TRUE

Current Behavior

Environment

win11 Cmake 3.29.6 CUDA 12.5 cuDNN 9.2

I have tested my CUDA installation in different ways (including testing in PyTorch). The CUDA works fine.

python setup.py install error

If I run python setup.py install, the terminal says the dlib is installed successfully. And the terminal logs:

-- Looking for cuDNN install...
-- Found cuDNN: C:/Program Files/NVIDIA/CUDNN/v9.2/lib/12.5/x64/cudnn.lib
-- Enabling CUDA support for dlib.  DLIB WILL USE CUDA, compute capabilities: 50
-- Configuring done (15.7s)
-- Generating done (0.5s)

However, the dilb is NOT indeed installed successfully.

When I use import dlib in a .py file, the terminal shows error: ImportError: DLL load failed while importing _dlib_pybind11: not found, as is described in #2977

pip install dlib --verbose error

If I run pip install dlib --verbose, the terminal says CUDA was found but your compiler failed to compile a simple CUDA program so dlib isn't going to use CUDA.

However, the dlib is indeed installed.

I can use import dlib in a .py file without error. But dlib.DLIB_USE_CUDA returns false.

Steps to Reproduce

run pip install dlib --verbose

or

run python setup.py install

following http://dlib.net/compile.html

Anything else?

I have tested using Python 3.7/3.8/3.9/3.10/3.11/3.12 and dlib 19.24.0/19.24.1/19.24.2/19.24.3/19.24.4, all these tests come with the same problem.

davisking commented 3 months ago

Hard to say, but something about your compiler or cuda install is broken. Which is out of control of dlib.

Do this to find out more:

cd dlib/cmake_utils/test_for_cuda/
mkdir build
cd build
cmake ..
cmake --build .

and see why that test program fails to build. It's just a trivial cuda program so you have to figure out why your computer is not capable of building it. All dlib's installer is doing is running that test build and if it fails it prints that message about not being able to use cuda.

eddiehe99 commented 3 months ago

Nothing seems to go wrong.

PS C:\Users\Eddie\Downloads\dlib-19.24.3> cd dlib/cmake_utils/test_for_cuda/
PS C:\Users\Eddie\Downloads\dlib-19.24.3\dlib\cmake_utils\test_for_cuda> mkdir build

    目录: C:\Users\Eddie\Downloads\dlib-19.24.3\dlib\cmake_utils\test_for_cuda

Mode                 LastWriteTime         Length Name
----                 -------------         ------ ----
d-----          2024/8/3     14:59                build

PS C:\Users\Eddie\Downloads\dlib-19.24.3\dlib\cmake_utils\test_for_cuda> cd build
PS C:\Users\Eddie\Downloads\dlib-19.24.3\dlib\cmake_utils\test_for_cuda\build> cmake ..
-- Building for: Visual Studio 17 2022
-- The C compiler identification is MSVC 19.40.33812.0
-- The CXX compiler identification is MSVC 19.40.33812.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: C:/Program Files/Microsoft Visual Studio/2022/Community/VC/Tools/MSVC/14.40.33807/bin/Hostx64/x64/cl.exe - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: C:/Program Files/Microsoft Visual Studio/2022/Community/VC/Tools/MSVC/14.40.33807/bin/Hostx64/x64/cl.exe - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
CMake Warning (dev) at CMakeLists.txt:10 (find_package):
  Policy CMP0146 is not set: The FindCUDA module is removed.  Run "cmake
  --help-policy CMP0146" for policy details.  Use the cmake_policy command to
  set the policy and suppress this warning.

This warning is for project developers.  Use -Wno-dev to suppress it.

-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.5 (found suitable version "12.5", minimum required is "7.5")
-- Configuring done (14.5s)
-- Generating done (0.1s)
-- Build files have been written to: C:/Users/Eddie/Downloads/dlib-19.24.3/dlib/cmake_utils/test_for_cuda/build
PS C:\Users\Eddie\Downloads\dlib-19.24.3\dlib\cmake_utils\test_for_cuda\build> cmake --build .
适用于 .NET Framework MSBuild 版本 17.10.4+10fbfbf2e

  1>Checking Build System
  Building NVCC (Device) object CMakeFiles/cuda_test.dir/Debug/cuda_test_generated_cuda_test.cu.obj
  cuda_test.cu
  cuda_test.cu
  tmpxft_00004358_00000000-10_cuda_test.cudafe1.cpp
  Building Custom Rule C:/Users/Eddie/Downloads/dlib-19.24.3/dlib/cmake_utils/test_for_cuda/CMakeLi
  sts.txt
  CMake is re-running because C:/Users/Eddie/Downloads/dlib-19.24.3/dlib/cmake_utils/test_for_cuda/
  build/CMakeFiles/generate.stamp is out-of-date.
    the file 'C:/Users/Eddie/Downloads/dlib-19.24.3/dlib/cmake_utils/test_for_cuda/build/CMakeFiles
  /cuda_test.dir/cuda_test_generated_cuda_test.cu.obj.depend'
    is newer than 'C:/Users/Eddie/Downloads/dlib-19.24.3/dlib/cmake_utils/test_for_cuda/build/CMake
  Files/generate.stamp.depend'
    result='-1'
  CMake Warning (dev) at CMakeLists.txt:10 (find_package):
    Policy CMP0146 is not set: The FindCUDA module is removed.  Run "cmake
    --help-policy CMP0146" for policy details.  Use the cmake_policy command to
    set the policy and suppress this warning.

  This warning is for project developers.  Use -Wno-dev to suppress it.

  -- Configuring done (0.3s)
  -- Generating done (0.3s)
  -- Build files have been written to: C:/Users/Eddie/Downloads/dlib-19.24.3/dlib/cmake_utils/test_
  for_cuda/build
  cuda_test.vcxproj -> C:\Users\Eddie\Downloads\dlib-19.24.3\dlib\cmake_utils\test_for_cuda\build\D
  ebug\cuda_test.lib
  Building Custom Rule C:/Users/Eddie/Downloads/dlib-19.24.3/dlib/cmake_utils/test_for_cuda/CMakeLi
  sts.txt
PS C:\Users\Eddie\Downloads\dlib-19.24.3\dlib\cmake_utils\test_for_cuda\build>
davisking commented 3 months ago

You didn't run cmake --build .

eddiehe99 commented 3 months ago

The output of cmake --build . is:

PS C:\Users\Eddie\Downloads\dlib-19.24.3\dlib\cmake_utils\test_for_cuda\build> cmake --build .
适用于 .NET Framework MSBuild 版本 17.10.4+10fbfbf2e

  1>Checking Build System
  Building NVCC (Device) object CMakeFiles/cuda_test.dir/Debug/cuda_test_generated_cuda_test.cu.obj
  cuda_test.cu
  cuda_test.cu
  tmpxft_00001e98_00000000-10_cuda_test.cudafe1.cpp
  Building Custom Rule C:/Users/Eddie/Downloads/dlib-19.24.3/dlib/cmake_utils/test_for_cuda/CMakeLi
  sts.txt
  cuda_test.vcxproj -> C:\Users\Eddie\Downloads\dlib-19.24.3\dlib\cmake_utils\test_for_cuda\build\D
  ebug\cuda_test.lib
  Building Custom Rule C:/Users/Eddie/Downloads/dlib-19.24.3/dlib/cmake_utils/test_for_cuda/CMakeLi
  sts.txt
davisking commented 3 months ago

Hard to say what's wrong with your system. But do pip uninstall dlib and ensure it is really uninstalled. Then do python setup.py install and see if it works now.

eddiehe99 commented 3 months ago

Unfortunately, the same error mentioned occurs.

ImportError: DLL load failed while importing _dlib_pybind11
StealUrKill commented 2 months ago

Unfortunately, the same error mentioned occurs.

ImportError: DLL load failed while importing _dlib_pybind11

I commented on a closed issue of the same thing. I happened to find the fix.

I found the fix for this. At least on windows for me.

https://stackoverflow.com/questions/62255730/dlib-importerror-in-windows-10-on-line-dlib-pybind11-import-dll-load-failed

This can be solved by copying the cudnn64_7.dll (available here: https://developer.nvidia.com/cudnn) into the %CUDA_PATH%/bin directory (probably something like this: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin)

I am using the latest from below

Visual Studio 2022 Cmake TensorRT 10.4 GA for Windows 10, 11, Server 2019, Server 2022 and CUDA 12.0 to 12.6 ZIP Package cuDNN 9.4.0 CUDA Toolkit 12.6 Update 1

Install the CUDA toolkit first and reboot. After install the cuDNN. The path should have these already in it from the install C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\bin C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\libnvvp C:\Program Files\NVIDIA\CUDNN\v9.4\bin\ C:\Program Files\NVIDIA\CUDNN\v9.4\

With system variables like these CUDA_PATH C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6 CUDA_PATH_V12_6 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6

Now in the directory C:\Program Files\NVIDIA\CUDNN\v9.4

Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\bin\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\bin Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\include\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\inlcude Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\bin\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\bin

If you're using tensorrt as well then you want to extract the zip and move all files and folders to C:\Program Files\NVIDIA\CUDNN\v9.4

Then you want to copy the file like the stackoverflow says (for me it was this)

This can be solved by copying the cudnn64_9.dll (from C:\Program Files\NVIDIA\CUDNN\v9.4\bin) into the %CUDA_PATH%/bin directory (probably something like this: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\bin)

Then

git clone https://github.com/davisking/dlib.git cd dlib mkdir build cd build cmake .. -DDLIB_USE_CUDA=1 -DUSE_AVX_INSTRUCTIONS=1 -DCMAKE_PREFIX_PATH="C:/Program Files/NVIDIA/CUDNN/v9.4"

You should see

-- Looking for cuDNN install...

-- Found cuDNN: C:/Program Files/NVIDIA/CUDNN/v9.4/lib/cudnn.lib -- Building a CUDA test project to see if your compiler is compatible with CUDA... CMake Warning (dev) at C:/Users/user/Downloads/dlib-master/dlib-master/dlib/cmake_utils/test_for_cuda/CMakeLists.txt:10 (find_package): Policy CMP0146 is not set: The FindCUDA module is removed. Run "cmake --help-policy CMP0146" for policy details. Use the cmake_policy command to set the policy and suppress this warning.

This warning is for project developers. Use -Wno-dev to suppress it.

-- Building a cuDNN test project to check if you have the right version of cuDNN installed... CMake Warning (dev) at C:/Users/user/Downloads/dlib-master/dlib-master/dlib/cmake_utils/test_for_cudnn/CMakeLists.txt:7 (find_package): Policy CMP0146 is not set: The FindCUDA module is removed. Run "cmake --help-policy CMP0146" for policy details. Use the cmake_policy command to set the policy and suppress this warning.

This warning is for project developers. Use -Wno-dev to suppress it.

-- Enabling CUDA support for dlib. DLIB WILL USE CUDA, compute capabilities: 50 -- Configuring done (12.2s) -- Generating done (0.1s) -- Build files have been written to: C:/Users/user/Downloads/dlib-master/dlib-master/dlib/build

Then cmake --build . --config Release and then you should see

-- Looking for cuDNN install... -- Found cuDNN: C:/Program Files/NVIDIA/CUDNN/v9.4/lib/cudnn.lib -- Enabling CUDA support for dlib. DLIB WILL USE CUDA, compute capabilities: 50 -- Configuring done (0.6s) -- Generating done (0.2s)

Run the following command from the source directory

python setup.py install

and then another output like

-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.6 (found suitable version "12.6", minimum required is "7.5") -- Looking for cuDNN install... -- Found cuDNN: C:/Program Files/NVIDIA/CUDNN/v9.4/lib/cudnn.lib -- Building a CUDA test project to see if your compiler is compatible with CUDA... CMake Warning (dev) at C:/Users/user/Downloads/dlib-master/dlib-master/dlib/cmake_utils/test_for_cuda/CMakeLists.txt:10 (find_package): Policy CMP0146 is not set: The FindCUDA module is removed. Run "cmake --help-policy CMP0146" for policy details. Use the cmake_policy command to set the policy and suppress this warning.

This warning is for project developers. Use -Wno-dev to suppress it.

-- Building a cuDNN test project to check if you have the right version of cuDNN installed... CMake Warning (dev) at C:/Users/user/Downloads/dlib-master/dlib-master/dlib/cmake_utils/test_for_cudnn/CMakeLists.txt:7 (find_package): Policy CMP0146 is not set: The FindCUDA module is removed. Run "cmake --help-policy CMP0146" for policy details. Use the cmake_policy command to set the policy and suppress this warning.

This warning is for project developers. Use -Wno-dev to suppress it.

-- Enabling CUDA support for dlib. DLIB WILL USE CUDA, compute capabilities: 50 -- Configuring done (23.3s) -- Generating done (0.1s) -- Build files have been written to: C:/Users/user/Downloads/dlib-master/dlib-master/build/temp.win-amd64-cpython-311/Release Invoking CMake build: 'cmake --build . --config Release -- /m' MSBuild version 17.10.4+10fbfbf2e for .NET Framework

A successful build then ends with

Installed c:\users\user\appdata\local\programs\python\python311\lib\site-packages\dlib-19.24.99-py3.11-win-amd64.egg Processing dependencies for dlib==19.24.99 Finished processing dependencies for dlib==19.24.99

After all is said and done

Python 3.11.9 (tags/v3.11.9:de54cf5, Apr  2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import dlib
>>> print(dlib.DLIB_USE_CUDA)
True

Screenshot 2024-09-11 001249

eddiehe99 commented 1 month ago

Unfortunately, the same error mentioned occurs.

ImportError: DLL load failed while importing _dlib_pybind11

I commented on a closed issue of the same thing. I happened to find the fix.

I found the fix for this. At least on windows for me.

https://stackoverflow.com/questions/62255730/dlib-importerror-in-windows-10-on-line-dlib-pybind11-import-dll-load-failed

This can be solved by copying the cudnn64_7.dll (available here: https://developer.nvidia.com/cudnn) into the %CUDA_PATH%/bin directory (probably something like this: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin)

I am using the latest from below

Visual Studio 2022 Cmake TensorRT 10.4 GA for Windows 10, 11, Server 2019, Server 2022 and CUDA 12.0 to 12.6 ZIP Package cuDNN 9.4.0 CUDA Toolkit 12.6 Update 1

Install the CUDA toolkit first and reboot. After install the cuDNN. The path should have these already in it from the install C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\bin C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\libnvvp C:\Program Files\NVIDIA\CUDNN\v9.4\bin\ C:\Program Files\NVIDIA\CUDNN\v9.4\

With system variables like these CUDA_PATH C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6 CUDA_PATH_V12_6 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6

Now in the directory C:\Program Files\NVIDIA\CUDNN\v9.4

Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\bin\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\bin Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\include\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\inlcude Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\bin\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\bin

If you're using tensorrt as well then you want to extract the zip and move all files and folders to C:\Program Files\NVIDIA\CUDNN\v9.4

Then you want to copy the file like the stackoverflow says (for me it was this)

This can be solved by copying the cudnn64_9.dll (from C:\Program Files\NVIDIA\CUDNN\v9.4\bin) into the %CUDA_PATH%/bin directory (probably something like this: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\bin)

Then

git clone https://github.com/davisking/dlib.git cd dlib mkdir build cd build cmake .. -DDLIB_USE_CUDA=1 -DUSE_AVX_INSTRUCTIONS=1 -DCMAKE_PREFIX_PATH="C:/Program Files/NVIDIA/CUDNN/v9.4"

I highly appreciate your detailed steps! It works!

Based on my different combinations of configurations, these steps are crucial after the installation of CUDA and cuDNN:

  1. Move files from subfolders to parent folders.

    Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\bin\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\bin Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\include\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\inlcude Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\lib\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\lib

  2. Specify the CMAKE_PREFIX_PATH Run the code below:

    git clone https://github.com/davisking/dlib.git cd dlib mkdir build cd build cmake .. -DDLIB_USE_CUDA=1 -DUSE_AVX_INSTRUCTIONS=1 -DCMAKE_PREFIX_PATH="C:/Program Files/NVIDIA/CUDNN/v9.4"

cmake --build . --config Release

python setup.py install

But actually, I did not do this. I did the following:

Firstly, I added the system variable CMAKE_PRIFIX_PATH whose value is "C:/Program Files/NVIDIA/CUDNN/v9.4".

Secondly, reboot.

Thirdly, run 'python setup.py install' from the source directory.

I searched the internet and found that the bizarre error may be related to the upgrade of Nvidia's cuDNN. When users install cuDNN 8.x - 1.x which are provided as zip files, the tutorials existing on the internet tell users to move the files in the bin, include, lib folders of cuDNN to the corresponding bin, include, lib folders of CUDA. However, the cuDNN (Graphical Installation) greater than 9.0.0 will create a subfolder such as 12.6. And the system variables automatedly generated do not lead to the right filepath. I tried to configure the system variables. It failed.

Consequently, I think the Tarball installation guide and your suggestions are correct.

StealUrKill commented 1 month ago

Unfortunately, the same error mentioned occurs.

ImportError: DLL load failed while importing _dlib_pybind11

I commented on a closed issue of the same thing. I happened to find the fix. I found the fix for this. At least on windows for me. https://stackoverflow.com/questions/62255730/dlib-importerror-in-windows-10-on-line-dlib-pybind11-import-dll-load-failed This can be solved by copying the cudnn64_7.dll (available here: https://developer.nvidia.com/cudnn) into the %CUDA_PATH%/bin directory (probably something like this: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin) I am using the latest from below Visual Studio 2022 Cmake TensorRT 10.4 GA for Windows 10, 11, Server 2019, Server 2022 and CUDA 12.0 to 12.6 ZIP Package cuDNN 9.4.0 CUDA Toolkit 12.6 Update 1 Install the CUDA toolkit first and reboot. After install the cuDNN. The path should have these already in it from the install C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\bin C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\libnvvp C:\Program Files\NVIDIA\CUDNN\v9.4\bin\ C:\Program Files\NVIDIA\CUDNN\v9.4\ With system variables like these CUDA_PATH C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6 CUDA_PATH_V12_6 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6 Now in the directory C:\Program Files\NVIDIA\CUDNN\v9.4 Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\bin\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\bin Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\include\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\inlcude Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\bin\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\bin If you're using tensorrt as well then you want to extract the zip and move all files and folders to C:\Program Files\NVIDIA\CUDNN\v9.4 Then you want to copy the file like the stackoverflow says (for me it was this) This can be solved by copying the cudnn64_9.dll (from C:\Program Files\NVIDIA\CUDNN\v9.4\bin) into the %CUDA_PATH%/bin directory (probably something like this: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\bin) Then git clone https://github.com/davisking/dlib.git cd dlib mkdir build cd build cmake .. -DDLIB_USE_CUDA=1 -DUSE_AVX_INSTRUCTIONS=1 -DCMAKE_PREFIX_PATH="C:/Program Files/NVIDIA/CUDNN/v9.4"

I highly appreciate your detailed steps! It works!

Based on my different combinations of configurations, these steps are crucial after the installation of CUDA and cuDNN:

  1. Move files from subfolders to parent folders.

Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\bin\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\bin Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\include\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\inlcude Move all the 12.6 files from C:\Program Files\NVIDIA\CUDNN\v9.4\lib\12.6 to C:\Program Files\NVIDIA\CUDNN\v9.4\lib

  1. Specify the CMAKE_PREFIX_PATH Run the code below:

git clone https://github.com/davisking/dlib.git cd dlib mkdir build cd build cmake .. -DDLIB_USE_CUDA=1 -DUSE_AVX_INSTRUCTIONS=1 -DCMAKE_PREFIX_PATH="C:/Program Files/NVIDIA/CUDNN/v9.4"

cmake --build . --config Release

python setup.py install

But actually, I did not do this. I did the following:

Firstly, I added the system variable CMAKE_PRIFIX_PATH whose value is "C:/Program Files/NVIDIA/CUDNN/v9.4".

Secondly, reboot.

Thirdly, run 'python setup.py install' from the source directory.

I searched the internet and found that the bizarre error may be related to the upgrade of Nvidia's cuDNN. When users install cuDNN 8.x - 1.x which are provided as zip files, the tutorials existing on the internet tell users to move the files in the bin, include, lib folders of cuDNN to the corresponding bin, include, lib folders of CUDA. However, the cuDNN (Graphical Installation) greater than 9.0.0 will create a subfolder such as 12.6. And the system variables automatedly generated do not lead to the right filepath. I tried to configure the system variables. It failed.

Consequently, I think the Tarball installation guide and your suggestions are correct.

Yeah I am knew to all this and I found out if making the python whl you dont need to do the cmake. But I guess it does not hurt lol

dlib-issue-bot commented 1 week ago

Warning: this issue has been inactive for 35 days and will be automatically closed on 2024-11-19 if there is no further activity.

If you are waiting for a response but haven't received one it's possible your question is somehow inappropriate. E.g. it is off topic, you didn't follow the issue submission instructions, or your question is easily answerable by reading the FAQ, dlib's official compilation instructions, dlib's API documentation, or a Google search.

blvz commented 1 week ago

I had similar problems and couldn't fix them just by moving the files to parent directories. In the end, what worked for me, was importing os and adding CUDA and CUDNN paths to BOTH PATH and DLL directories:

if 'ON' == 'ON':
    import os
    os.environ["PATH"] += os.environ["PATH"] + ";" + r"C:\\Program Files\\NVIDIA\\CUDNN\\v9.4\\bin\\12.6" + ";" + r"C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.6\\bin"
    os.add_dll_directory('C:\\Program Files\\NVIDIA\\CUDNN\\v9.4\\bin\\12.6')
    os.add_dll_directory('C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.6\\bin')

and for anyone struggling with this, you can edit this file in site-packages, example: .venv\Lib\site-packages\dlib\__init__.py
or open the .whl with 7zip and edit it there, example: dlib-19.24.6-cp312-cp312-win_amd64.whl\dlib\__init__.py
(just don't forget to force reinstall the wheel if you do the latter!)