Untitled project

I created a basic transformer model for the classification task and started working on training it.

I need to think about what augmentations to use on the training set.

Also I need to find a way to balance the dataset (I am using the FMA medium dataset). The problem is that there is a class (Rock music) that has much more samples, than all of the other classes combined. I need to find a way, to balance this, because right now the model learns only to always predict the rock class.

I downloaded the dataset and made a simple dataset class. I got rid of empty or broken samples and made a graph which genre / number of samples

I am working on an AI model, that given the songs waveform or spectrogram, can detect what genre it belongs to.