Equivariance Loss not converging on custom data

[guozixunnicolas]

Hello,

I appreciate the authors' effort in making the open-source code available and presenting it SOOO neatly!

One question regarding the equivariance loss, I trained PESTO from scratch using my custom data. But the eqv loss is not converging properly (see screenshot below). It still seems to work during evaluation though.

I wonder whether this this common? Or should I tweak the hyperparameters to reduce the eqv loss further.

Many thanks and looking forward to your reply!

Best,
Nicolas