OpenCV导入DLL加载失败及Debug编译报错求助（Win10编译含contrib）

Hey there, let's break down your two main issues—Debug build failure and the ImportError: DLL load failed—using your specific environment details (Win10, NVIDIA P5000, CUDA 9.0, VS2015, Python 3.6.4 non-Anaconda).

1. Fixing Debug Build Failure #

Debug builds often trip up on configuration mismatches or dependency handling that Release mode ignores. Here's what to check:

• CUDA Debug Library Configuration
  Make sure CMake is set to link against CUDA's Debug runtime libraries. Verify these CMake flags:

  • Set CUDA_USE_STATIC_CUDA_RUNTIME to OFF (unless you specifically need static linking)
  • Ensure CUDA_CUDA_LIBRARY points to the Debug version of cudart (usually in C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\lib\x64\cudart_static.lib for Debug)
    After adjusting, re-run CMake and regenerate the VS project.

• Contrib Module Debug Dependencies
  Some contrib modules (like cudaarithm, cudafeatures2d) have stricter Debug mode requirements. Check the VS error log to see which module is failing—if it's a contrib module, make sure:

  • You've downloaded the full contrib source matching OpenCV 3.4.1 (no missing files or mismatched versions)
  • CMake didn't skip any required dependencies for that module in Debug mode (look for lines like Skipping module X due to missing dependency in the CMake output)

• VS2015 Debug Compile Settings
  Match OpenCV's runtime library setting to your project:

  • In VS, go to Project Properties > C/C++ > Code Generation > Runtime Library
  • Set it to Multi-threaded Debug DLL (/MDd)—this should align with how you configured OpenCV in CMake (check CMAKE_CXX_FLAGS_DEBUG to confirm)

2. Resolving ImportError: DLL load failed #

This error almost always boils down to missing dependencies or path mismatches. Let's tackle it:

• Use Release-Built cv2.pyd with Standard Python
  Important note: The standard Python 3.6.4 you installed is a Release build. Debug versions of cv2.pyd (from your failed Debug build) won't work here—only the Release-built cv2.pyd is compatible. Make sure you're copying the right file:

  • The Release cv2.pyd is usually in <build_dir>\lib\Release
  • Copy it to your Python's site-packages folder (e.g., C:\Python36\Lib\site-packages)

• Verify Dependency DLLs
  Use the VS Command Prompt to check what DLLs cv2.pyd needs:

  dumpbin /dependents C:\Python36\Lib\site-packages\cv2.pyd
  

  Look for missing DLLs:

  • OpenCV Release DLLs: Ensure <build_dir>\bin\Release is added to your system PATH (or copy those DLLs to the same folder as cv2.pyd)
  • CUDA DLLs: cudart64_90.dll (from C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin) must be in your PATH
  • VS 2015 Runtime DLLs: msvcp140.dll and vcruntime140.dll should be present (install the VS 2015 Redistributable if missing)

• Double-Check CMake Python Configuration
  When you ran CMake, confirm these paths were set correctly (no typos or wrong versions):

  • PYTHON3_INCLUDE_DIR: C:\Python36\include
  • PYTHON3_LIBRARY: C:\Python36\libs\python36.lib
  • PYTHON3_EXECUTABLE: C:\Python36\python.exe
    If any were wrong, re-run CMake, regenerate the project, and recompile the Release build.

内容的提问来源于stack exchange，提问作者Alex