Initial Model trained on individual actions

The first iteration of the model. 20 videos per action, 30 frames each. Uses hand and pose landmarks. Issues with stability and accuracy.

Model trained on a sequence of actions

Tried training on sequences of actions to improve context understanding, too many frames caused lag. Even with an idle for a transition it still struggled to tell the difference between two different sequences of actions.

Better model with sequence checking

This model has 26 actions with a high rate of accuracy and stability. It can can check for sequences of actions and tells the user which sequence was done correctly. Only have issuees with back fist actions and outside inside blocks being misclassified.

Model with early feedback integration

This is the same model as the previous video, but with early feedback integration. It has some feedback on what actions it is expecting next based on the sequence. With some issues that it is only giving feedback for one of the the three sequences. And it is not very specific or with a lot of detail.