Nividia jetson nano

the model does not seem to work well. The joints are pretty random and the number of detected people in this photo is 2. I used the weights you provided in the example. are they accurate ?

The model provided should work acurate enough.
Where are you trying this in RobotIgniteAccademy? Take into account that in the accademy you are not using GPU/CUDA, so the performance is worse than it will be in the Jetson Nano card. The accademy examples are ment only to show the code and how it works, but if you want good performance you will need to :

  1. Excute the code in GPU enabled instances in ROSDevelopementStudio: ROSDS
  2. Execute the code in a physical JetsonNano as explained in the course. You will see that the performance gets multiplied by 10.