I\'m using a pretrained model with output (1,1,1920) as a layer for my image denoising network with input (32,32,3). The pretrained model is DenseNet201 without the top layer so