From aac932ed845f55a2590bdbab410e53108259eda3 Mon Sep 17 00:00:00 2001 From: AndReGeist Date: Thu, 18 Jul 2024 16:22:31 +0200 Subject: [PATCH] Update README.md Fixed typo in readme --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 6b5e993..e5d7fd4 100644 --- a/README.md +++ b/README.md @@ -35,7 +35,7 @@ While the *choice of loss function* is important for learning with rotations, we - **For rotation estimation (rotations in model output)** use $\mathbb{R}^9+\mathrm{SVD}$ or $\mathbb{R}^6+\mathrm{GSO}$. If the regression targets are only small rotations, using quaternions with a halfspace-map is a good option. -- **For feature prediction (rotations in model input)** use $\mathbb{R}^9+\mathrm{SVD}$ or $\mathbb{R}^6+\mathrm{GSO}$. If under memory constraints, quaternions with a halfspace-map and data-augmentation are viable. +- **For feature prediction (rotations in model input)** use rotation matrices. If under memory constraints, quaternions with a halfspace-map and data-augmentation are viable. > [!NOTE] > To support these recommendations, we conducted several experiments and reproduced the results of previous works.