Training on Piano Roll data about bicyclegan HOT 6 CLOSED

I am not familiar with the piano roll data. Is it for future prediction? The current generator architectures assume strong spatial alignment between input and output, which might not be the case of your application. It also seems that L1 reconstruction does not work for your application as most of the pixels are black.

from bicyclegan.

Yes, it is for future prediction. Does the spatial alignment come from the concatenation of features?
With other architectures L1 didn't seem to help...

from bicyclegan.

Our generator architecture works well for spatially aligned data. It might be struggling with other types of data.

from bicyclegan.

What in your generator's architecture makes it work well for spatially aligned data?

from bicyclegan.

The generator is a U-Net with many skip connections. Please see the original paper for more details. Also, check out Sec 3.2.1 in the pix2pix paper.

from bicyclegan.

Thank you for pointing out the section in the paper. I'll close the issue.

from bicyclegan.