Recommended OpenCV version, CMake flags, or pre-built binaries for CUDA 13.2 & cuDNN 9.2

C++
,
1

I am planning to use OpenCV with CUDA support and I would like some advice on version compatibility, the build process, or alternative solutions.

My current environment setup is:

  • CUDA Toolkit: 13.2 (correctly installed and added to the system PATH)

  • cuDNN: 9.2 (correctly installed)

  • OS: Windows 10

  • GPU: NVIDIA RTX A4000

I have a few questions to help me get started:

  1. Version Compatibility: Which branch or specific release of OpenCV (and opencv_contrib) is currently recommended and tested to fully support CUDA 13.2 and cuDNN 9.2?

  2. Build Process & Flags: What are the essential and recommended CMake flags I need to configure for this setup?

  3. Comprehensive Guide: Does anyone know of a reliable, step-by-step tutorial or updated documentation covering the build process from source for these recent versions?

  4. Pre-compiled Binaries: Alternatively, are there any trusted community repositories, wheels, or pre-built binaries available that already include CUDA 13.2 and cuDNN 9.2 support, allowing me to skip the manual build process?

Any advice, examples of CMake commands, or pointers to specific repositories would be greatly appreciated.

Thank you in advance for your help!

2

I myself am unaware of any tables that state which CUDA works with which opencv_contrib. I myself don’t deal with CUDA, so that information might exist somewhere.

We have, or used to have, a CUDA specialist on the forum. He hasn’t been active since ~November.

You might be able to glean some information from this: