L4T (Linux for Tegra) Multimedia API with ffmpeg, jetson-ffmpeg, and opencv4. This is a lightweight portable API for hardware-accelerated encoding/decoding on NVIDIA Jetson.

1. Dependencies

• jetson-ffmpeg: https://github.com/jocover/jetson-ffmpeg
• opencv4: https://docs.opencv.org/4.x/d7/d9f/tutorial_linux_install.html

2. nvmpi-portable

git clone https://github.com/jheo4/nvmpi-portable.git
cd nvmpi-portable && mkdir build
cd build && cmake ..
make -j$(nproc) && sudo make install

3. testing

g++ nvmpi_test.cc `pkg-config --libs --cflags libavcodec libavformat libavutil opencv4 nvmpi `
g++ portable_test.cc `pkg-config --libs --cflags libavcodec libavformat libavutil opencv4 nvmpi nvmpi_portable`

It's for HW-accelerated encoding/decoding at the code level and each frame granularity. jetson-ffmpeg is easily usable with ffmpeg, but there are some glitches using it with libavcodec. It seems because of the discrepancies between libavcodec contexts and jetson-ffmpeg. So, instead of using nvmpi via libavcodec interfaces, nvmpi-portable directly use nvmpi for encoding/decoding. nvmpi-portable has dependencies of nvmpi, ffmpeg, and opencv4 because nvmpi-portable delegates the image/video-related compelxities to them. Moreover, it is opencv compatible for better usability.

So, it can be easily usable as a library and integratble to other projects; I wrote this to use for my research project :). You can compare the compelxities between nvmpi_test.cc and portable_test.cc.

• nvmpi-portable is inspired by jetson-ffmpeg and I refer it.
• L4T API: https://docs.nvidia.com/jetson/l4t-multimedia/index.html