Hi,
I am trying to use fastai for my master thesis where I use a Unet to get morphological information about bacterial cells. Hence I dont directly use the output of the Unet but rather extracts the features at the head which later is used for some unsupervised clustering. I am currently at a stage where I would like to see how the skip connections are affecting the feature maps that I am extracting, meaning I would like to drop all/some of the skip connections. I have without luck tried to do so with use of hooks. Are there any easy way to play with the skip connections or do I have to build the encoder/decoder myself - Not that I want to be lazy, I just wanted to see if something already exists within the fastai enviroment. I hope some of you have some input.
Best,
Michael
It is utilized for tasks that the prediction has the same spatial dimension as the input such as image segmentation, optical flow estimation, video prediction, etc. By introducing skip connections in the encoder-decoded architecture, fine-grained details can be recovered in the prediction.
}–>RapidFS
Thanks for the answer Paris 
I do understand the concept behind the skip connections, but I would like to see the actual effect by removing these - just for testing. That is why I was curious if this is supported in Fastai, but it doesn’t seems like it is.