At the moment I am writing my bachelor thesis on optical music recognition using Deep Learning. I started dealing with fastai yesterday (just finished the first lecture of part1). If I understand it correctly, fastai is a highly abstract library that helps to deal with Pytorch. I’m pretty new to the field of Deep Learning so that comes handy, followed some tutorials on CNN and basic tensorflow stuff. But I don’t feel able to set up a Autoencoder filtering staves from sheet music and identify the notes within these staves with tensorflow.
Fastai sounds more like click and play.
I am currently reading a paper and would like to recreate the architecture. Can I just put the single layers on top of each other?
@Tgtrmr
Welcome to the community!
Great! All the best for your thesis. I don’t want to discourage you, but it seems very unlikely that you’d be able to create the architecture you want after watching just the first lesson, especially since you are only a beginner in Deep Learning. I’d personally recommend you continue with this course. Towards the end of the course, you’ll have a better understanding of how to tackle this situation and what steps you’d need to take to build your model.
Even to use pre-built code, you need to understand a few technicalities- about whats exactly going in, and coming out of your model, and whats going on inside the model. Without knowing a few things, it becomes almost impossible to make any changes, or to even fix errors!
Cheers, stay safe!
Definitely. Personally, I’ve taken a few courses before fastai. And fastai was the one that enabled me to do stuff on my own - end to end. So yeah, please continue! All the best!