Skip to content

install.md

Latest commit

 

History

History
153 lines (103 loc) · 5.56 KB

install.md

File metadata and controls

153 lines (103 loc) · 5.56 KB

Environment Installation Guide

📌 Install CMake

Important Notes
  • Installing CMake serves two main purposes: compiling OpenCV and building your own projects.
  • While project compilation has no strict version requirements, compiling OpenCV does. Using incompatible versions may cause errors.
  • The version number provided here is a reference based on project compatibility.

1. Version Selection

cmake version 3.15.7

2. Official Download and Installation

Download Link. Select version cmake-3.15.7-win64-x64.msi. After downloading, run the installer and note the installation path.

Notes
  • Other versions on the same page are also acceptable.

📌 Install CUDA

Important Notes
  • While CUDA is likely already installed, note that installing CUDA before Visual Studio may cause compilation failures in VS Code.
  • CUDA libraries for Windows C++ programs require cl.exe (the Visual Studio compiler). The VS compiler version must match CUDA's requirements.
  • CUDA 11.6 requires VS2019, whose official standalone installer is no longer available.
  • Installing CUDA after VS may cause unrecognized library errors.

1. Install Visual Studio 2019

Download via link. During installation, select components as shown below:

Installation Configuration

2. Install CUDA & cuDNN

  • Follow standard tutorials. During CUDA installation, choose Custom Installation and select all available options (except pre-installed drivers).
  • After installation, verify that ~\bin and ~\lib paths are added to system environment variables.
Notes
  • Selecting all options ensures necessary compute libraries are included.
  • bin contains precompiled binaries – missing PATH entries cause runtime errors (e.g., "missing xx file").
  • lib contains dynamic libraries – missing PATH entries cause linking errors during execution.
  • OpenCV and TensorRT have similar dependencies.

📌 Install OpenCV

Important Notes
  • OpenCV offers precompiled binaries or self-compiled versions. Precompiled versions (e.g., opencv_world) are simpler but lack CUDA acceleration.
  • This guide focuses on compiling OpenCV with CUDA support.
  • Successful compilation requires zero errors; otherwise, the install step will fail.

1. Download OpenCV Source

  1. Download from official site. Select version 4.5.5 and choose Sources.

OpenCV Download

2. Compile OpenCV with CMake

a. Initial Configuration

  • Launch C:\Program Files\CMake\bin\cmake-gui.exe.
  • Configure as shown below:
    1. Source code path (OpenCV folder).
    2. Build directory (new empty folder).
    3. Specify generator: Visual Studio 2019.
  • Click Configure → Generate.

OpenCV Compilation

b. Second-Round Configuration

Why Reconfigure?
  • Initial configuration often fails due to missing dependencies.
  • CUDA acceleration requires the opencv_contrib module.

Fix Common Errors

  1. Python mismatch: Disable Python-related options by searching PYTHON in CMake and unchecking all entries.
  2. Download failures: Manually download missing 3rd-party libraries:
    • Visit OpenCV 3rdparty.
    • Locate the cache folder (e.g., C:\opencv_s\opencv-4.5.5\.cache\ffmpeg).
    • Download the required branch (e.g., ffmpeg/src), rename files to match cache hashes, and replace them.

Add Essential Modules

  1. Search CUDA and enable all CUDA-related options.
  2. Download opencv_contrib 4.5.5.
  3. Set OPENCV_EXTRA_MODULES_PATH to the extracted opencv_contrib/modules path.
  4. Disable unused modules (e.g., wechat_qrcode, xfeatures2d) if they cause errors.

Final Configuration

  1. Re-run Configure → Generate. Resolve any remaining errors.
  2. Ensure no red entries appear before proceeding.

3. Generate Install Files with Visual Studio 2019

  1. Click Open Project in CMake to launch VS2019.
  2. Navigate to Build → Batch Build. Check INSTALL under Debug or Release.
  3. Compilation takes ~4 hours.
  4. If minor errors occur, manually deselect problematic modules. After successful build:
    • Add ~\install\x64\vc16\bin and ~\install\x64\vc16\lib to system PATH.

OpenCV Generation

📌 Install TensorRT

Download version TensorRT-10.12.0.36 from the official site. Extract the package, then add ~\TensorRT-10.12.0.36\bin and ~\TensorRT-10.12.0.36\lib to system PATH.

📌 Install LibTorch

  1. Download libtorch-win-shared-with-deps-1.13.0+cu116.zip from this link.
  2. Extract the package, then add bin and lib folders to system PATH.

All dependencies are now configured.