Differences between the different backward convolutions ?


I was wondering what were the difference between the backward data and the backward filter convolutions. It is not very clear what the dimensions of each part should be in the documentation.

I guess the “data” is to compute an image back from the features and the “filter” is to compute the filter gradients from the errors. Is that correct ?

Is there a doc with some examples as to when and with what kind of parameters each function should be used ?