Just sharing. I tried many things and now get LB 0.96 from a single model all the time. Ensembling does not work!!! Mixing two or more models, each with 0.96, does not make it better! It’s very confusing!
A theoretical discussion: what does it mean if different encoders get the same result?
One version: any reasonable encoder is good enough! The payload is in the decoder?
Hi @jc138691
That’s a great way to look at it. If different encoders are getting similar results that likely means that the encoder will do most of the heavy lifting in the denoising process.
ThinkOnward Team