pytorch-metric-learning icon indicating copy to clipboard operation
pytorch-metric-learning copied to clipboard

Published 20 hours ago • verified publisher icon KevinMusgrave

Spherical Embedding Constraint (SEC)

Open StefanoSalvatori opened this issue 3 years ago • 2 comments

Do you plan to add the Spherical Embedding Constraint (SEC) proposed in the following paper? (https://arxiv.org/pdf/2011.02785.pdf)

StefanoSalvatori avatar Jun 28 '21 07:06 StefanoSalvatori

Is it possible that SEC is equivalent to this already-implemented regularizer?

https://kevinmusgrave.github.io/pytorch-metric-learning/regularizers/#centerinvariantregularizer

I can't find a public version of the paper referenced in the docs at that link, but per the discussion in section 3.3 of the paper you linked (in particular the paragraph following equation 17) I think using centerinvariantregularizer with the default L2 distance is equivalent to SEC.

dmarx avatar Sep 06 '21 03:09 dmarx

For reference, here's a screenshot from the "center invariant" paper:

image

KevinMusgrave avatar Sep 06 '21 19:09 KevinMusgrave