如何在Windows（GTX 1050）的Anaconda环境中安装支持GPU加速的OpenCV？

Got it, let's break this down—you're dealing with a super common pain point: getting GPU-accelerated OpenCV to play nice with Anaconda, instead of only working in the global Python CMD environment. Here are two reliable methods to get this sorted, plus tips to boost your YOLO detection speed once it's working.

Method 1: Use Conda-Forge's Precompiled GPU-Ready OpenCV (Easiest Path) #

This is the fastest route, since conda-forge maintains prebuilt OpenCV packages with CUDA support tailored specifically for Anaconda environments.

• First, activate your target Anaconda environment:

  conda activate your_env_name
  

• Uninstall any existing OpenCV packages to avoid conflicts (cover both pip and conda-installed versions):

  pip uninstall opencv-python opencv-contrib-python -y
  conda uninstall opencv -y
  

• Install OpenCV with CUDA support. Your GTX 1050 uses Pascal architecture (compute capability 6.1), which works best with CUDA 11.x (CUDA 12+ drops support for Pascal GPUs). Install the matching cudatoolkit alongside OpenCV:

  conda install -c conda-forge opencv-contrib-python cudatoolkit=11.8
  

• Verify success: Launch Python in your activated environment and run:

  import cv2
  print(cv2.cuda.getCudaEnabledDeviceCount())
  

  If it returns 1 (or more, if you have multiple GPUs), you're good to go!

Method 2: Compile OpenCV from Source (For Customization) #

If the precompiled package doesn't fit your needs (e.g., you need specific extra modules), compiling from source gives you full control.

Prerequisites #

• Ensure your NVIDIA driver is up to date (for CUDA 11.8, you need driver version ≥450.80.02)
• Install dependencies in your Anaconda environment:

  conda install cmake numpy visual-studio-build-tools  # Use VS Build Tools or MinGW on Windows
  

• Download matching versions of OpenCV source and OpenCV_contrib (e.g., both v4.8.0)

Compilation Steps #

1. Create a build folder inside your OpenCV source directory
2. Open CMake GUI, set the source directory to your OpenCV root, and build directory to the new build folder
3. Click "Configure" and select your compiler (e.g., Visual Studio 2022 for Windows)
4. Enable these critical options:
   • WITH_CUDA (check this to enable GPU support)
   • WITH_CUDNN (enable if you want faster DNN inference via CuDNN)
   • OPENCV_EXTRA_MODULES_PATH (point to the modules folder in your OpenCV_contrib download)
   • CUDA_ARCH_BIN (set to 6.1 to match your GTX 1050's compute capability)
   • CMAKE_INSTALL_PREFIX (set to your Anaconda environment's root folder, e.g., C:\Users\YourName\anaconda3\envs\your_env_name)
5. Click "Configure" again until all red lines disappear, then click "Generate"
6. Open the generated .sln file in Visual Studio, build the INSTALL project in Release mode
7. After compilation, OpenCV files will be installed directly into your Anaconda environment

Verify Compilation #

Same as Method 1: run cv2.cuda.getCudaEnabledDeviceCount() in your activated environment—you should see a non-zero number.

Bonus: Boost YOLO Detection Speed #

Once GPU-accelerated OpenCV is working, make sure your YOLO code actually utilizes the GPU:

import cv2

# Load YOLO model
net = cv2.dnn.readNet("yolov3.weights", "yolov3.cfg")
# Force DNN backend to use CUDA
net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)
net.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA)

This should push your FPS way above 7—you'll likely see a 5-10x improvement with your GTX 1050.

内容的提问来源于stack exchange，提问作者BaoLocPham