Currently I am trying to use this framework for image recognition, but after a while I get the exeption "Error during claculation". The sgd trainer increased the gradients until the gradients of one layer reached neagtive infintity as far as I saw.
Projekt_NN_Conv_Net.zip