Another edition of BigLiminal using the same dataset as Klein, this time for Ernie.

I've never used Ernie, so I'll play with the images later.

So far it seems to be the only model that actually learned BigLiminal correctly. Almost every other model fails.

For the Klein edition dataset, the one I'm using here, I hand tagged about 400 or so of the 950 images with corrections and added details. I think that helped a lot with this version.

The dataset is available for MIT use.

Have at it.