Open fantes opened 5 years ago
There is special hardware(NVJPEG) for scaling images and it uses 0 CPU and GPU https://developer.nvidia.com/nvjpeg https://docs.nvidia.com/jetson/l4t-multimedia/l4t_mm_jpeg_encode.html https://docs.nvidia.com/jetson/l4t-multimedia/l4t_mm_jpeg_decode.html
input connector 's resize is done only on CPU, and takes more time then inference on many GPU enabled platforms.
When available, cv::resize should be replaced by cv::cuda::resize of cv::gpu::resize (depending on opencv version)
BIG FAT WARNING: this needs a GPU enabled opencv, which is not the case on default ubuntu, neither on nvidia's jetpack (!??! : see https://devtalk.nvidia.com/default/topic/1029172/jetpacks-opencv-with-cuda/)
notes : compiling opencv whith cuda support for TX2 jetpack : https://jkjung-avt.github.io/opencv3-on-tx2/