search

vcg-uvic / learned-correspondence-release

Code release for "learning to find good correspondences" CVPR 2018
Other
258 stars 54 forks source link

Normalisation of the ground-truth smallest eigenvector #15

Open ivan-pavlov opened 5 years ago

ivan-pavlov commented 5 years ago

Hello!

Thank you for the code release for "Eigendecomposition-free Training of Deep Networks with Zero Eigenvalue-based Losses" paper.

In the paper, you desribe a normalisation of the ground-truth smallest eigenvector:

image

I can see how you are building the ground-truth smallest eigenvector for the loss computation here.

But, unfortunately, I could not find in the loss function or elsewhere in the code how you scale and translate this eigenvector.

Could you please clarify to me how you perform the normalisation of the ground-truth smallest eigenvector?

kmyi commented 4 years ago

@Dangzheng Can you reply? I guess we are quite late on replying though.