I tested Hyperpose on Jetson Xavier NX and I get only 1.5 FPS for Openpose-Coco Model but the repo tells that hyperpose is 4 times faster than openpose.I tested the same on my laptop and hyperpose is faster than openpose.I couldn’t replicate the same on xavier NX.Am I missing out on anything? . I am using tensorrt optimized models
Have you maxmized the device performance first?
$ sudo nvpmodel -m 0 $ sudo jetson_clocks
I have a similar issue…I only get 1.5fps and the device is set to maximum performace
Sorry that we don’t have too much experiment on hyperpose.
Maybe you can check if they have optimized for the embedded system first.
For Jetson, we also have an example to run an openpose model with TensorRT.
We can get 251 fps with a resnet18 based pose estimation model on the Xavier.
Please check the following GitHub for the details: