[FEATURE REQUEST] On Jetson Nano: Warn about OpenCV 4.1.1 cmake not finding cuDNN when using JetPack 4.4

[TeddybearCrisis]

Describe the issue OpenCV 4.1.1 cmake doesn't find cuDNN when using JetPack 4.4

System

• Jetson Nano
• cmake 3.16.3
• L4T 32.4.3 [ JetPack 4.4 ]
  • Ubuntu 18.04.5 LTS
  • Kernel Version: 4.9.140-tegra
• CUDA 10.2.89
  • CUDA Architecture: 5.3
• OpenCV version: 4.1.1
  • OpenCV Cuda: NO
• CUDNN: 8.0.0.180
• TensorRT: 7.1.3.0
• Vision Works: 1.6.0.501
• VPI: 0.3.7

To Reproduce

1. Download and install the Jetson Nano SD Card image https://developer.nvidia.com/jetson-nano-sd-card-image JP 4.4 from 2020/07/07
2. optional: install cmake 3.16.3
3. Checkout this repo and run: ./buildOpenCV.sh |& tee openCV_build.log In the cmake config you can see that cuDNN -> NO is diplayed and OpenCV is built without it

Expected behavior cuDNN -> YES is detected by cmake and OpenCV is built with cuDNN

Additional context I was using a newer version of cmake as my build pipline requires this. So there's no issue related reason for the updated cmake version. This is likely not the cause for the error.

[TeddybearCrisis]

Explicitely providing the flag -D CUDNN_VERSION='8.0' resolved the issue. Via https://github.com/mdegans/nano_build_opencv Maybe modify the shell script with a little switch statement selecting the explicit verison depending on the JetPack version?

[bryanwux]

Did you solve this issue? I am working on building OpenCV 4.3.0 using JetPack4.4 on Xavier Agx and meet the same problem.

[TeddybearCrisis]

Well it is solved as described in my previous comment. Yet I wouldn't call this a proper solution but a workaround. I'd say it's up to the maintainer to provide a fix or classify my comment as solution.

[bryanwux]

@TeddybearCrisis Thank you! I use this method and my OpenCV 4.4.0 manages to find cuDNN!