search

lzhengning / SubdivNet

Subdivision-based Mesh Convolutional Networks.
MIT License
251 stars 34 forks source link

Dataset #11

Closed tt-study closed 3 years ago

tt-study commented 3 years ago

Why does each sample in the 'cubes' dataset have 10 subdivision samples in the 'Cubes-MAPS-48-4'? Are all the 10 subdivision samples trained as training samples? Thank you

lzhengning commented 3 years ago

Hi, @tt-study ,

Why does each sample in the 'cubes' dataset have 10 subdivision samples in the 'Cubes-MAPS-48-4'

Because randomness is introduced in the MAPS algorithm, for different remeshing of a shape. We generated multiple subdivision samples, as a way of data augmentation. This is effective when the training dataset is small.

Are all the 10 subdivision samples trained as training samples.

Yes.