Open steb6 opened 1 year ago
Now the model can work with more (or less) than 5 actions and with a variable number of examples for each class. Here there is an example of the current support set, where I added some of the actions cited above;
During this modification, I also discovered the following issue: https://github.com/hsp-iit/workbook-stefano-berti/issues/3
In the latest experiment, this is the support set:
this is the inference result:
In this moment, the action inside the Support Set (and so the ones that we are able to recognize) are the following:
It would be nice to recognize other actions, but we need to consider that at the moment the Action Recognition module works with 3D skeletons that do not contain information about objects or hands. However, we try with some other actions.
Everyone can add here additional actions that would be nice to have:
High five (could be distinguished from hello thanks to the temporal context)
Rock paper scissor (the robot could detect that the person wants to play a match and could do chose something random among rock paper or scissor)
Stop (doing cross with both arms on the chest could make the robot stop what he is doing)
Dab (the robot could detect dab and could do the same