Hi everyone,
starting from ResNet and UNet implementation in FastAI I’d like to implement the FuseNet network. This network for semantic segmentation consists of two parallel encoders which contribute were summed together.
Where can I start? Is there a way to use Resnet and Unet in FasAI as building blocks for a more complex model?
Thank you to all
Hello and welcome gmaker.
I do not know where you are starting from, so please excuse any false assumptions about what you already know and don’t yet know.
fastai is built on PyTorch, and the way to assemble fastai’s models together is with PyTorch’s Module class. So I suggest studying PyTorch’s Module class, tutorials, and these forums for examples of assembling complex models.
You will also need to understand how to bring a PyTorch model into fastai and how to use fastai with it. Again, the forums, and the video lessons.
A possible shortcut is to search the web, especially Github, for implementations that others have already done. Often someone else has already done the work to translate a paper’s code into various frameworks. If you find a PyTorch version of FuseNet, you are almost done. You can bring it into fastai directly as is, or substitute fastai’s building blocks for some of its components.
HTH to get you started. Good luck with your project! Malcolm 