I'm trying to convert a fairly complicated convolutional neural network from Caffe to Caffe2 for performance reasons. I am aware of the tutorial for converting equivalent layers, however, my DummyData layers in Caffe crash the conversion. While I know that I will most likely have to extend the tutorial script to account for this, I was hoping that someone could point me in the correct direction for constructing the equivalent of this type of layer in Caffe2.
I also have to convert a couple of different python layers, so it's pretty inevitable that I will need to modify the tutorial in the Caffe2 Github repo; I'm just not sure how to approach this and thought that seeing how the DummyData layer is done would be very helpful. Unfortunately, I don't have access to the code I have written, as it is on my computer at work, however, all I have done so far is added a new function with similar args to those given to all the other layers and created a new operator. I'm not sure if this is the correct approach (as there may be a largely equivalent existing operator that I missed).
