FC-DCNN: A densely connected neural network for stereo estimation

Abstract

We propose a novel lightweight network for stereo estimation. Our network consists of a fully-convolutional densely connected neural network (FC-DCNN) that computes matching costs between rectified image pairs. Our FC-DCNN method learns expressive features and performs some simple but effective postprocessing steps. The densely connected layer structure connects the output of each layer to the input of each subsequent layer. This network structure and the fact that we do not use any fullyconnected layers or 3D convolutions leads to a very lightweight network. The output of this network is used in order to calculate matching costs and create a cost-volume. Instead of using time and memory-inefficient cost-aggregation methods such as semiglobal matching or conditional random fields in order to improve the result, we rely on filtering techniques, namely median filter and guided filter. By computing a left-right consistency check we get rid of inconsistent values. Afterwards we use a watershed foreground-background segmentation on the disparity image with removed inconsistencies. This mask is then used to refine the final prediction. We show that our method works well for both challenging indoor and outdoor scenes by evaluating it on the Middlebury, KITTI and ETH3D benchmarks respectively. Our

A Stereo Matching Algorithm for High‐Precision Guidance in a Weakly Textured Industrial Robot Environment Dominated by Planar Facets

“…Dominik et al . [HF21] proposed a lightweight fully convolutional densely connected neural network (FC‐DCNN) for stereo estimation. Cheng et al .…”

A binocular reconstruction based on perspective projection constraints and its application on robot eye‐hand coordination

“…Guo et al [23] proposed construction of the cost volume by group-wise correlation to enhance performance. Dominik et al [24] proposed a lightweight fully convolutional densely CNN for stereo estimation. Poggi et al [25] introduced weakly guided stereo matching, leveraging a small number of sparse yet reliable depth measurements retrieved from an external source for training.…”

“…Dominik et al. [24] proposed a lightweight fully convolutional densely CNN for stereo estimation. Poggi et al.…”

FCDSN-DC: An Accurate and Lightweight Convolutional Neural Network for Stereo Estimation with Depth Completion

“…The method FC-DCNN [3] by D. Hirner and F. Fraundorfer is used as a baseline implementation for this method. There, it has been shown that replacing the feature extraction step with a deep learning approach in order to learn highly-dimensional and expressive features already outperforms traditional stereo methods such as SGM.…”

“…In this work we extend upon this method in two ways: Adding a network structure to learn a better similarity function between image patches and adding a network structure that learns to complete the sparse disparity map instead of using handcrafted post-processing steps. We end up with a fully trainable method that outperforms the baseline method of FC-DCNN [3] in all evaluated datasets and is comparable with other state-of-the art deep-learning methods. The whole method split into every step is illustrated in Fig.…”