MT3 LODO GuitarSet

[Drew Edwards]

In the original MT3 paper, you discuss leave-one-dataset-out / zero-shot experiments. 

I'm a bit confused on the results, specifically the reported metric for GutiarSet.

https://openreview.net/pdf?id=iMSjopcOn0p

Here is the LODO table (Table 5). Below it is the zero-shot table (Table 6). It looks like most of the entries from Table 6 come from Table 5 for the case when the left-out dataset is equal to the evaluation dataset. Is this correct?

If so, why is GuitarSet reported for the case where MAESTRO is left out? Shouldn't the result be 0.32, where GuitarSet is both the hold-out and the evaluation dataset?

Sorry if this has been pointed out already. If I am misunderstanding the reported results, can you please clarify?

[Ian Simon]

+Josh Gardner 

It looks you are correct!  I always wondered why that GuitarSet number was so high :)

-Ian

--
Magenta project: magenta.tensorflow.org
To post to this group, send email to magenta...@tensorflow.org
To unsubscribe from this group, send email to magenta-discu...@tensorflow.org
---
To unsubscribe from this group and stop receiving emails from it, send an email to magenta-discu...@tensorflow.org.