Saiteja Utpala
Saiteja Utpala
@nguigs what do you think
@lpereira95, almost true. Here is the example for which 1 tangent vec, multiple base points makes sense. `SPDMatrices`. $T_{X}\text{SPD}(n) = \\{X\\} \times \text{SYM}(n) \equiv \text{SYM}(n)$. I think (one tangent vec,...
Just to list out manifold properties. @ninamiolane @nguigs @ythanwerdas * Euclidean : Connected : True, Compact : False * General Linear : Connected : True, Compact: False (general linear as...
@Ryan-Vaughn . Hi Thanks for looking into this. most of it is done. But you can take over PR and get it merged!
you need to use gs backend for computations not specific backend (like you use numpy).
I see. But if user changes tolerance `torch.set_default_dtype(torch.float64)` behaviour is not dependent ? we would still be returning based float32
@lpereira95 tagging some references [Region Covariance: A Fast Descriptor for Detection and Classification](https://www.porikli.com/mysite/pdfs/porikli%202006%20-%20Region%20Covariance%20A%20Fast%20Descriptor%20for%20Detection%20and%20Classification.pdf). what I have done is just take one global covariance descriptor of single image. But we can...
@lpereira95 , No hurry. Take your time!
@ninamiolane @lpereira95 Github actions is working fin, see #1469 (just checking sometimes it does't trigger)
@yuxiangw. Thanks! one more question if we want to pick best leanring rate (by picking best training error achieved) privately. Is there any reference ?