Yes. That can be done in at least 2 ways:
gstreamer, you would access your CSI cam with
nvarguscamerasrc, and you would be able to use CUDA processing on frames in NVMM memory with
nvivafilter plugin that allows to build a custom library for your processing.
With jeston MMAPI. You may install it (
sudo apt install nvidia-l4t-jetson-multimedia-api) and chek the examples installed into
No, above ways would be much easier.
It is the same physical memory. What has already been allocated before for Linux,X,Ubuntu… would of course not be available.
In older L4T releases there may have been some limitations to the amout of memory that could be allocated for CUDA/GPU, but I think that now most of it is available. Someone more skilled may correct me if wrong.