DeMoN: Depth and Motion Network for Learning Monocular Stereo
DeMoN: Depth and Motion Network for Learning Monocular Stereo
复制标题
DOI：
10.1109/cvpr.2017.596
复制
发表时间：
2016-12
期刊：
影响因子：
--
通讯作者：
Benjamin Ummenhofer;Huizhong Zhou;J. Uhrig;N. Mayer;Eddy Ilg;Alexey Dosovitskiy;T. Brox
中科院分区：
文献类型：
--
作者：
Benjamin Ummenhofer;Huizhong Zhou;J. Uhrig;N. Mayer;Eddy Ilg;Alexey Dosovitskiy;T. Brox
在本文中，我们将运动结构定义为一个学习问题。我们训练一个端到端的卷积网络，从连续的、无约束的图像对中计算深度和相机运动。该架构由多个堆叠的编码器-解码器网络组成，核心部分是一个能够改进自身预测的迭代网络。该网络不仅估计深度和运动，而且还估计表面法线，图像之间的光流和匹配的置信度。该方法的一个关键组成部分是基于空间相对差异的训练损失。与传统的两帧结构自运动方法相比，该方法的结果更准确，鲁棒性更强。与流行的单图像深度网络相比，DeMoN学习了匹配的概念，因此可以更好地推广到训练过程中看不到的结构。
In this paper we formulate structure from motion as a learning problem. We train a convolutional network end-to-end to compute depth and camera motion from successive, unconstrained image pairs. The architecture is composed of multiple stacked encoder-decoder networks, the core part being an iterative network that is able to improve its own predictions. The network estimates not only depth and motion, but additionally surface normals, optical flow between the images and confidence of the matching. A crucial component of the approach is a training loss based on spatial relative differences. Compared to traditional two-frame structure from motion methods, results are more accurate and more robust. In contrast to the popular depth-from-single-image networks, DeMoN learns the concept of matching and, thus, better generalizes to structures not seen during training.