In lesson 7, Jeremy mentions that densenets are nicely suited for segmentation problems. However, in lesson 3 when Jeremy uses a U-net with a ResNet encoder, it does as well as the Tiramisu network, which is a Densenet-based segmentation model that has a U-net structure. Is there something I am missing regarding the application of Densenets to segmentation? Is the ResNet model only comparable to Tiramisu because of the many training tricks fastai uses?