KieranMcAuliffe Hi all, I'm by no means a computer vision expert, but I'm trying to figure out on the OAK-D how to run faster pose estimation rivaling that of the Kinect. There are lots of papers out there on how to do this, but I'm hoping to find a well maintained prebuilt solution that I can easily integrate. I assume this would involve using the depth image. -K
jakaskerl KieranMcAuliffe https://hub.luxonis.com/ai/models/12acd8d7-25c0-4a07-9dff-ab8c5fcae7b1?view=page is a pose estimation model based on the yolov8n - it can run at about 33FPS on the OAK-D Thanks, Jaka
KieranMcAuliffe Thanks Jaka, I believe this is only (x, y) coordinates and not (x, y, z) correct? Generally, is there an easy way to figure out what the output tensors represent for models in the Hub?
jakaskerl KieranMcAuliffe The decoding is done automatically for this model so you don't have to worry about it. The output is X, Y, but you can overlay it over depth so it gives you the Z component as well. Thanks, Jaka
