After installing the latest Jetpack (4.4) yesterday on Xavier NX, OpenCV face_recognition has become unstable and extremely slow. I suspect GPU acceleration no longer works. If true, how can I install OpenCV with GPU Acceleration? Any insight/comment on this issue is greatly appreciated.
Regards
Hi,
Please try to re-install OpenCV with the script and try again:
https://github.com/AastaNV/JEP/blob/master/script/install_opencv4.3.0_Jetson.sh
It enables -D WITH_CUDA=ON. May be required for face recognition with GPU acceleration.
The current script seems to turn off cuDNN due to v8.0 API change issue. Is there a current script to build OpenCV with GPU Acceleration on Jetson Xavier NX Jetpack 4.4 with CUDA and cuDNN enabled?
How do I enable CUDA for OpenCV with the current Jetpack 4.4 image from Nvida?
You can try the script by community member @mdegans
GitHub - mdegans/nano_build_opencv: Build OpenCV on Nvidia Jetson Nano
That script seems to have CUDNN_VERSION=‘8.0’ as well as the one from https://github.com/AastaNV/JEP/blob/master/script/install_opencv4.3.0_Jetson.sh. According to the change log “turn off cuDNN due to v8.0 API change issue” it seems that both can’t be on. Am I reading that correctly?
It appears that running the GitHub - mdegans/nano_build_opencv: Build OpenCV on Nvidia Jetson Nano script may break cuDNN.
I’m currently running the versions that are part of the Jetpack 4.4 image and concerned I may break something if I run either script. Doe anyone know what issues are meant by the v8.0 API change issue?