Dynamic UNet Reference/Description

Hey Forums!

For a scientific publication, we are using the fast.ai DynamicUNet implementation that is provided within the libary, and we use a ResNet18 encoder to train our data set.

However, I am struggling to find a good citable source for this approach. Is there any good resource that we can reference when describing our model? We can of course describe what it is doing, but based on what I remember from the course there is also some more twists to it (some pixel shuffle for upsampling I think?) and we don’t want the reviewers to think that we came up with this.

I have seen this thread What is Dynamic Unet and the referenced repo, but I don’t think that is from the original author?

How do other people approach this problem of referencing (parts of) the fastai library?


I went back through the course materials, and the only thin I can find is jeremy saying they didn’t write paper yet about all the smart things that make this work…

@jeremy did you ever publish anything along those lines or put up an arxiv paper?
I also found your arxiv paper about fastai v2 but it seems that is only a technical documentation of the library itself?