How to describe dataset labels for action recognition and prototxt by using N-dim Convolution

[根岸儀和]

Hi.

I am interested in action recognition, and I am intended to make dataset for

training my own action recognition neural net models by using N-dimension convolution(i.e. 3D convolution).

But, there is few text for Caffe's implementation of N-dim convolution,

and I neither know how to describe dataset labels which applies for action recognition,

nor understand well how to describe training/testing(validiating) prototxt.

Does anyone have ideas?

Thanks.