the problem about the rot linear -- x4_reshape[:, 0]

[Cherished-l]

hi i can't understand about below: why the rotation and contrastive mlp choosed x4_reshape[:, 0] and x4_reshape[:, 1]? why is 0 and 1? why do u choose the patch?

[yanchaoz]

the same question

[tangy5]

This should be a sampling option, derived and adopted from the rotation prediction and reconstruction tasks. It can be other feature indices or probabilistic if Bayesian is used.

Thanks.