Support for cv::gpu::VideoReader_GPU in L4T21.1

I can’t even construct this object using OpenCV4Tegra on L4T21.1, it just throws_nogpu. Was this compiled correctly? It looks like a very promising object since it reads into a cv::gpu::GpuMat&.

ShervinE@NVIDIA, do you know when, or if, the VideoReader_GPU will be supported in OpenCV4Tegra?

I’m not sure but I’ll look into it and get back to you.

The GPU video decoding in OpenCV4Tegra uses NVCUVID library, and unfortunately this isn’t available for Jetson TK1, so OpenCV4Tegra on Jetson TK1 was built without GPU video decoding. (OpenCV4Tegra on Jetson TK1 also doesn’t have GPU accelerated rendering).

VisionWorks will come with both GPU accelerated video input and output rendering on Jetson TK1. (I can’t comment on when VisionWorks will be publicly available). But if you need GPU accelerated decoding before then, you will need to write your own code to input the video stream to OpenCV using Gstreamer, since Gstreamer is GPU accelerated on Jetson TK1.