(Reference PyTorch Code: https://github.com/Shape-Lab/SPHARM-Net)

https://towardsdatascience.com/geometric-deep-learning-for-spherical-data-55612742d05f

https://www.lead-dbs.org/helpsupport/knowledge-base/atlasesresources/cortical-atlas-parcellations-mni-space/

• Rotational Equivariance → 적절히 회전했을 때 값이 똑같은 것

• Rotatonal Invariance → 회전과 관계 없이 값 자체가 완벽히 똑같은 것

Spherical CNNs

https://frost-crate-a82.notion.site/Spherical-CNNs-181f7e966c0e8012b88ad1bce7c23ffe?pvs=4

SPHARM-Net

Rotation Equivariant Convolutional Filter + Spectral Convolution

→ Enable effective tessellation without predefined neighborhoods or data augmentation