Hey, I am working with multi-class brain tumor segmentation and somebody suggested I use the fast.ai library for some tricks. I have my own unet which i am training from scratch. If i want to use the code in in https://github.com/fastai/fastai/blob/master/courses/dl2/carvana-unet.ipynb
where the learner gets defined as:
m = to_gpu(Unet34(m_base))
models = UnetModel(m)
learn = ConvLearner(md, models)
Here im guessing I can define my own UnetModel class and then what if i dont want to use pre-trained weights as im assuming is happening when Unet34 is being called. Can I just load my own weights or what?
Also if i want to feed in my own np array as dataset(cause it is in nii format) can I do that? or maybe enter my own pytorch dataset loader?
Thanks for your help!