Making the caffe model rotationally variant (Imagenet)

I am trying to train a caffe model using the Imagenet example that is rotationally variant. Does anybody know which parameters in train_val.txt need to be changed to do so?

[xxw345]

What you mean parameters to change? Could you detail it?

I meant, which modifications do I need to make to the network in train_val to make the model rotationally variant? I am trying to train a model that allows me to classify images based on orientation

[xxw345]

I don't have experience on this. My experience always rotate the samples in training set to increase the ability of rotation invariant. I am guessing if give the different rotation (orientation) with different labels (one class in end fully connected layer) will help?