Ankur Handa: Quaternions and Euler angles are discontinuous and difficult for neural networks to learn. They show 3D rotations have continuous representations in 5D and 6D, which are more suitable for learning.
i.e. regress two vectors and apply Graham-Schmidt (GS).
11 replies, 679 likes
Matt Miesnieks: So this is literally 6D AI ? :) @6d_ai
8 replies, 69 likes
Marc B. Reynolds: This keeps show up in my timeline, so:
Quat's are not discontinuous by any normal definition.
4 replies, 14 likes
Chris Choy: I really like this paper from Hao Li's group: On the Continuity of Rotation Representations in Neural Networks, CVPR19 :) https://arxiv.org/pdf/1812.07035.pdf
0 replies, 12 likes
Found on Jan 26 2020 at https://arxiv.org/pdf/1812.07035.pdf