Created by: jczaja

Issue to track a task of adding a pass that will fuse conv2d_transpose of mkl-dnn with batch_norm mkldnn op. This should benefit internal face model.