1 文章说明

方向：立体匹配

会议：CVPR2019

2 动机

However, existing methods only use features from plain convolution layers or a simple aggregation of multi-level features to calculate cost volume, which is insufficient because stereo matching requires discriminative features to identify corresponding pixels in rectified stereo image pairs. In this paper, we propose a unary features descriptor using multi-level context ultra-aggregation (MCUA), which encapsulates all convolutional features into a more discriminative representation by intra- and inter-level features combination.

提出了一个更好的结构来提取更好的特征

3 核心

Multi-level Context Ultra-Aggregation (MCUA) scheme which combines the features at the shallowest, smallest scale and deeper, larger scales using just “shallow” skip connections.

提出的框架：

提出的核心模块：

1 通过上面的分支提取全局特征

2 通过下面的分支提取局部特征

3 通过短连接来融合不同尺度的局部和全局特征

4 数据库

The Scene Flow datasets ： 训练 35454

                                                测试：4370

                                                   像素：1242

 KITTI2015/2012 ：   训练  200/194

                                       测试：200/195

                                                   像素：1242/375

5训练

优化器：Adam (Adaptive Moment Estimation)

batch: 8

maximum disparity (D):192 pixels

Scene Flow datasets:  70 epochs, 256 × 512 resolution

KITTI2015/2012 : 

6 实验