Currently the model is just trained for the three actions that you mentioned above. Now you could use your own action detection model with the sample which would require some modifications of the sample code.
Also, what additional actions did you have in mind?
Hello DongMing. To be able to support more than the 3 basic actions using a model under deployment_tools/intel_models (sitting, standing, raising hand) requires a newly trained model provided by Intel - so in other words, data collection and model retraining performed by the Intel OpenVino development team. So, today, we only support those aforementioned 3 actions. New actions will be supported in future releases of this sample, however.
There is nothing stopping you from using an open source model (i.e. perhaps a TensorFlow model zoo item), properly training it using hardware of your choice, pumping it through OpenVino model optimizer, and finally using the Smart Classroom Example as a reference on how to code it using OpenVino inference_engine but this is a lot of work !
Thank You for using OpenVino !
Can we get the actual caffe model. I wish to convert the model again using modal optimizer to a different resolution. Currently the default resolution is [1x3x400x680]].