In the vanilla implementation of U-net and Linknet models, they don't contain dropout layers, right? But if you choose efficientnetB3 or efficientnetB7 as the backbone, the model contains dropout layers in its encoder?
Additionally, is it weird for my graph to look like this? I added data augmentation and added a dropout layer of 0.3 just before the last layer of the model. EfficienetB3 is the backbone.
Hello,
In the vanilla implementation of U-net and Linknet models, they don't contain dropout layers, right? But if you choose efficientnetB3 or efficientnetB7 as the backbone, the model contains dropout layers in its encoder?
Additionally, is it weird for my graph to look like this? I added data augmentation and added a dropout layer of 0.3 just before the last layer of the model. EfficienetB3 is the backbone.
Thanks.