search

LIX-shape-analysis / SURFMNet

code and data associated with the paper "Unsupervised Deep Learning for Structured Shape Matching"
50 stars 7 forks source link

Question about the data provided by the given link #9

Open dongliangcao opened 2 years ago

dongliangcao commented 2 years ago

Dear Author,

Currently, I am working with functional map. However, I encounter a problem about the data provided by the given link that I want to hear your answer. I notice that in your other works for example, GeomFmaps. You provide the MATLAB code to compute the eigenvalues, eigenvectors and transposed eigenvectors. I notice that when the transposed eigenvector is computed. It will multiply with the mass matrix. While, in the link you provide for the SURFMNet, I notice the transposed eigenvector doesn't multiply with the mass matrix. I want to know which process is correct. I also notice the .off files in the link you provide in this repo is different than other link, like GeomFmaps. Could you give me some hints.

Thanks a lot.

Best regards, Dongliang

Not-IITian commented 2 years ago

Hi,

Both are correct :) in GeomFmap code, we compute the tranposed eigen vector and thus we need to multiply with mass matrix. For this code, transposed vectors are already precomputed and thus already multiplied with mass matrix before.

Not-IITian commented 2 years ago

Do you refer to the aligned dataset we provide? Aligned dataset was manually aligned in meshlab given original Faust remesh and Scape remesh.

dongliangcao commented 2 years ago

Hi,

Both are correct :) in GeomFmap code, we compute the tranposed eigen vector and thus we need to multiply with mass matrix. For this code, transposed vectors are already precomputed and thus already multiplied with mass matrix before.

Thanks for your reply. But when I check the eigenvectors and eigenvalues generated by MATLAB code provided in GeomFmap and the already precomputed one in SURFMNet. I think they differ a lot and I also check the .off mesh files between GeomFmap and SURFMNet, I think the vertices positions are also different.

Not-IITian commented 2 years ago

Hi,

Thanks for letting me know. This should not be the case except if GeomFmap did some unit normalization or something. I will double check after Cvpr deadline.can you also check this data: https://github.com/Not-IITian/Weakly-supervised-Deep-Functional-map

dongliangcao commented 2 years ago

Hi,

Thanks for letting me know. This should not be the case except if GeomFmap did some unit normalization or something. I will double check after Cvpr deadline.can you also check this data: https://github.com/Not-IITian/Weakly-supervised-Deep-Functional-map

Thanks for your reply. I use the MATLAB code provided in GeomFmap to run on the .off data in https://github.com/Not-IITian/Weakly-supervised-Deep-Functional-map and also the .off data provided by this repository. I get the same eigenvectors and eigenvalues as use the .off data provided by GeomFmap. But they are different compared with the eigenvectors and eigenvalues provided by this repository.

dongliangcao commented 2 years ago

Hi, Thanks for letting me know. This should not be the case except if GeomFmap did some unit normalization or something. I will double check after Cvpr deadline.can you also check this data: https://github.com/Not-IITian/Weakly-supervised-Deep-Functional-map

Thanks for your reply. I use the MATLAB code provided in GeomFmap to run on the .off data in https://github.com/Not-IITian/Weakly-supervised-Deep-Functional-map and also the .off data provided by this repository. I get the same eigenvectors and eigenvalues as use the .off data provided by GeomFmap. But they are different compared with the eigenvectors and eigenvalues provided by this repository.

I also check the python version code provided by https://github.com/pvnieo/SURFMNet-pytorch/tree/master/surfmnet. I think the results are the same as the MATLAB version in GeomFmap.