Multi-Dimensional Residual Dense Attention Network for Stereo Matching

Abstract

Very deep convolutional neural networks (CNNs) have recently achieved great success in stereo matching. It is still highly desirable to learn a robust feature map to improve ill-posed regions, such as weakly textured regions, reflective surfaces, and repetitive patterns. Therefore, we propose an endto-end multi-dimensional residual dense attention network (MRDA-Net) in this paper, focusing on more comprehensive pixel-wise feature extraction. Our proposed network consists of two parts: the 2D residual dense attention net for feature extraction and the 3D convolutional attention net for matching. Our designed 2D residual dense attention net uses a dense network structure to fully exploit the hierarchical features from preceding convolutional layers and uses residual network structure to fuse low-level structure information and high-level semantic information. The 2D attention module of the net aims to adaptively recalibrate channel-wise features to be more concerned about informative features. Our proposed 3D convolutional attention net further expands attention mechanism for matching. The stacked hourglass module of the net is exploited to extract multi-scale context information as well as geometry information. The novel 3D attention module of the net aggregates hierarchical sub-cost volumes adaptively instead of manually and then achieves a comprehensive recalibrated cost volume for more correct disparity computation. The experiments demonstrate that our approach achieves the state-of-the-art accuracy on Scene Flow dataset and KITTI 2012 and KITTI 2015 Stereo datasets. INDEX TERMS Stereo matching, multi-dimensional, residual dense attention, hierarchical, 3D attention mechanism.

A Survey on Theories and Applications for Self-Driving Cars Based on Deep Learning Methods

“…For example, Nguyen, et al [51] presented a network structure, which is constructed by a wide context learning network and stacked encoder-decoder 2D CNNs. Zhang, et al [52] proposed an end-to-end multidimensional residual dense attention network, which focuses on more comprehensive pixel-level feature extraction. The network includes a two-dimensional residual dense attention network for feature extraction and a three-dimensional convolutional attention network for matching.…”

MPA‐Net: multi‐path attention stereo matching network

“…The attention structure is an important composition of the computer vision networks, including semantic segmentation [31], person re‐identification [32, 33], depth estimation [34] and attention‐based stereo matching [35]. Specifically, SE‐Net [25] aims to improve the representation of first acquiring attention vector along the channel dimension, and then aggregating the features as the channel‐wise multiplication between input features and attention vector.…”

A target-free stereo matching method of 3D vision-based for measuring composite fiber bundles

“…However, fiber bundles with high reflectivity and weak texture characteristics suffer significant feature information loss during the imaging process, leading to reduced accuracy and an increased number of incorrect matching points during measurement. In response to computational challenges posed by stereo matching algorithms in extreme test environments, numerous scholars have undertaken extensive research, exploring classical matching algorithms, image pyramid optical flow methods, 3D-DIC methods, and deep learning-based approaches [11,12,13,14,15]. Nonetheless, these improved methods have not fully met the precision requirements for fiber bundle measurement.…”