论文标题
在透视上保留图像转换和模棱两可的刚性
Rigidity Preserving Image Transformations and Equivariance in Perspective
论文作者
论文摘要
我们表征了图像平面转换类别，这些变换实现了刚性摄像头动作，并将这些转换称为“保持刚度”。特别是，针孔图像的2D翻译不是保留刚度的。因此，当将CNN用于3D推理任务时，将电感偏差从均衡性转移到翻译到对刚性保留变换的刚性可能是有益的。我们研究了CNN中如何近似近似刚性的刚性，并在6D对象姿势估计和视觉定位上测试我们的想法。在实验上，我们改善了几个竞争基线。
We characterize the class of image plane transformations which realize rigid camera motions and call these transformations `rigidity preserving'. In particular, 2D translations of pinhole images are not rigidity preserving. Hence, when using CNNs for 3D inference tasks, it can be beneficial to modify the inductive bias from equivariance towards translations to equivariance towards rigidity preserving transformations. We investigate how equivariance with respect to rigidity preserving transformations can be approximated in CNNs, and test our ideas on both 6D object pose estimation and visual localization. Experimentally, we improve on several competitive baselines.
