Q: Is it possible to only use the palm tracking model of the Hand Landmarker?

[JC3]

Have I written custom code (as opposed to using a stock example script provided in MediaPipe)

Yes

OS Platform and Distribution

Ubuntu 22.04, macOS, Windows

MediaPipe Tasks SDK version

0.10.15

Task name (e.g. Image classification, Gesture recognition etc.)

Hand Landmarker

Programming Language and version (e.g. C++, Python, Java)

Python

Question

I am reading that the Hand Landmarker first uses a fast palm detector to locate the hand before running the full hand recognition model.

Is it possible (perhaps via clever confidence threshold settings?) to only use the palm detector and access the detected palm positions?

The reason I ask is we're using the Hand Landmarker to detect hands but we're running it on the CPU because no GPU is available (Intel NUC11). Therefore, it's got about 90-100ms of latency. However, we're only actually interested in the palm positions anyways, so I'm wondering if there's a way to just access the faster palm detector and use that instead to improve latency. We don't need the full 21-point hand model.

Thanks!