bmcfee / CDL

Convolutional dictionary learning and (sparse) coding
8 stars 4 forks source link

Where is the art_tatum-willow_weep_for_me.wav file? #5

Open hnyz979 opened 4 years ago

hnyz979 commented 4 years ago

Interested in your convolutional dictionary learning, and I want to read through your code. Could you tell me where is your data file? By the way, I found that your work was done in 2013. Nowadays, many deep learning modules such as tensorflow has been invented, is there any possibility of using tensorflow to implement CDL? Thank you~

bmcfee commented 4 years ago

Could you tell me where is your data file?

That file came from the cal500 dataset, but there was nothing special about it. I picked it mainly because it was solo piano, but you could substitute just about any other recording.

By the way, I found that your work was done in 2013. Nowadays, many deep learning modules such as tensorflow has been invented, is there any possibility of using tensorflow to implement CDL?

Sure, I don't see why not. The algorithm would be totally different though. The method here was following in the tradition of sparse coding / matrix factorization methods, and most of the work went into setting up the inference as a convex optimization problem that could be solved efficiently. It's probably not a great model to use though, since it's over-expressive and under-constrained in practice, so the resulting codes aren't terribly informative.