-2 opencv config_error ai_generated partial

cv2.error: OpenCV(4.7.0) /modules/dnn/src/dnn.cpp: error: (-2:Unspecified error) DNN module was not built with CUDA backend; switched to CPU

ID: opencv/dnn-cuda-backend-missing

Also available as: JSON · Markdown · 中文
75%Fix Rate
90%Confidence
1Evidence
2023-08-15First Seen

Version Compatibility

VersionStatusIntroducedDeprecatedNotes
4.5.0 active — — —
4.6.0 active — — —
4.7.0 active — — —
4.8.0 active — — —
4.9.0 active — — —

Root Cause

The OpenCV installation does not include CUDA support for DNN, so GPU acceleration is unavailable and falls back to CPU.

generic

中文

OpenCV 安装未包含 DNN 的 CUDA 支持，因此 GPU 加速不可用，回退到 CPU。

Official Documentation

https://docs.opencv.org/4.x/d2/d58/tutorial_dnn_googlenet.html

Workarounds

  1. 80% success Rebuild OpenCV from source with CUDA enabled: set -DWITH_CUDA=ON and -DWITH_CUDNN=ON in CMake, and ensure CUDA toolkit and cuDNN are installed. 
    Rebuild OpenCV from source with CUDA enabled: set -DWITH_CUDA=ON and -DWITH_CUDNN=ON in CMake, and ensure CUDA toolkit and cuDNN are installed.
  2. 70% success Use a pre-built OpenCV package that includes CUDA support, such as from conda-forge (opencv with cuda variant) or NVIDIA's official Docker images. 
    Use a pre-built OpenCV package that includes CUDA support, such as from conda-forge (opencv with cuda variant) or NVIDIA's official Docker images.

中文步骤

  1. Rebuild OpenCV from source with CUDA enabled: set -DWITH_CUDA=ON and -DWITH_CUDNN=ON in CMake, and ensure CUDA toolkit and cuDNN are installed.
  2. Use a pre-built OpenCV package that includes CUDA support, such as from conda-forge (opencv with cuda variant) or NVIDIA's official Docker images.

Dead Ends

Common approaches that don't work:

  1. Set environment variable OPENCV_DNN_CUDA=1 95% fail

    This environment variable does not exist; CUDA support must be compiled into OpenCV.

  2. Install CUDA toolkit after OpenCV installation 85% fail

    CUDA toolkit must be present at build time; installing it after does not add CUDA support to an existing OpenCV build.

  3. Use cv2.dnn.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA) to force CUDA 90% fail

    If the DNN module was not built with CUDA, this call will throw the same error; it does not enable CUDA at runtime.