About the time it takes for HandPose to recognize the video in the video?

[git-SteamedBun]

I use CMake to build HandPose and run it, but every frame captured by it takes nearly two seconds to run. How can I solve this?

[brmarkus]

Can you provide some description about your environment (SW and HW), please? Is the video maybe decoded using a SW-decoder because no HW-accelerated video decoding available in HW or no driver installed? Is the system load high (high CPU-load, high GPU-load) in the background? Have you build OpenCV on your own from source, or installed it via your OperatingSystem? Can I assume your OpenCV is using gstreamer as the backend for video decoding? Can you try to decode the video using e.g. a gstreamer command line in a terminal? Do you see the same latency?

[git-SteamedBun]

I'm very sorry, I didn't understand your explanation. My environment is Win10 and opencv is installed in the system, and I consume a lot of CPU when running HandPose, so do I need to modify the video decoder?