Holoscan SDK Question

I’m looking to capture video frames from the VideoStreamReplayerOp operator and manipulate the frames using opencv. My question is: Is there a Holoscan SDK method or gxf::Entity that allows me to capture the video frame data so that it can be changed by opencv?

Hi Tim, it sounds like you might want to create a native C++/Python operator in Holoscan that contains your custom OpenCV operations, and connect it to the output of VideoStreamReplayer. You can do this via tensor interop to and from Holoscan tensors. For Python, the documentation is here Creating Operators - NVIDIA Docs; for C++ Creating Operators - NVIDIA Docs. Some examples are on GitHub: holoscan-sdk/examples/tensor_interop at main · nvidia-holoscan/holoscan-sdk · GitHub, holoscan-sdk/examples/native_operator at main · nvidia-holoscan/holoscan-sdk · GitHub

Which language are you creating your application in?

1 Like

Did you mean the INPUT of the VidoStreamReplayer?

Could you elaborate on this? From your first post I’m understanding it as, you are replaying a saved video, and want to use OpenCV to modify the video frames.

The VideoStreamReplayerOp reads a file and plays that file, it is usually the first operator in an app if you’re replaying a saved file (vs streaming from a live source), so I did mean the output of this operator in the message above.

Got it. Thx!