MobileStereoNet: Towards Lightweight Deep Networks for Stereo Matching (2021-08-22T00:00:00.000000Z)

TL;DR

Two light models for stereo vision with reduced complexity and without sacrificing accuracy are proposed, based on a 2D and a 3D model with encoder-decoders built from2D and 3D convolutions, respectively.

Abstract

Recent methods in stereo matching have continuously improved the accuracy using deep models. This gain, however, is attained with a high increase in computation cost, such that the network may not fit even on a moderate GPU. This issue raises problems when the model needs to be deployed on resource-limited devices. For this, we propose two light models for stereo vision with reduced complexity and without sacrificing accuracy. Depending on the dimension of cost volume, we design a 2D and a 3D model with encoder-decoders built from 2D and 3D convolutions, respectively. To this end, we leverage 2D MobileNet blocks and extend them to 3D for stereo vision application. Besides, a new cost volume is proposed to boost the accuracy of the 2D model, making it performing close to 3D networks. Experiments show that the proposed 2D/3D networks effectively reduce the computational expense (27%/95% and 72%/38% fewer parameters/operations in 2D and 3D models, respectively) while upholding the accuracy. Code: https://github.com/cogsys-tuebingen/mobilestereonet.

Authors

A. Zell

4 papers

Faranak Shamsafar

1 papers

Samuel Woerz

1 papers

TL;DR

Abstract

Authors

References32 items

Separable Convolutions for Optimizing 3D Stereo Networks

NLCA-Net: a non-local context attention network for stereo matching

X3D: Expanding Architectures for Efficient Video Recognition

Point-GNN: Graph Neural Network for 3D Object Detection in a Point Cloud

“Looking at the Right Stuff” – Guided Semantic-Gaze for Autonomous Driving

Semantic Stereo Matching With Pyramid Cost Volumes

DeepPruner: Learning Efficient Stereo Matching via Differentiable PatchMatch

AutoDispNet: Improving Disparity Estimation With AutoML

GA-Net: Guided Aggregation Net for End-To-End Stereo Matching

Group-Wise Correlation Stereo Network

Fast Deep Stereo with 2D Convolutional Processing of Cost Signatures

Normalized Object Coordinate Space for Category-Level 6D Object Pose and Size Estimation

Occlusions, Motion and Depth Boundaries with a Generic Network for Disparity, Optical Flow or Scene Flow Estimation

3D Depthwise Convolution: Reducing Model Parameters in 3D Vision Tasks

CBMV: A Coalesced Bidirectional Matching Volume for Disparity Estimation

Pyramid Stereo Matching Network

MobileNetV2: Inverted Residuals and Linear Bottlenecks

Learning for Disparity Estimation Through Feature Constancy

Learning Spatio-Temporal Representation with Pseudo-3D Residual Networks

Look Wider to Match Image Patches With Convolutional Neural Networks

Cascade Residual Learning: A Two-Stage Convolutional Neural Network for Stereo Matching

SGM-Nets: Semi-Global Matching with Neural Networks

MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

End-to-End Learning of Geometry and Context for Deep Stereo Regression

Efficient Deep Learning for Stereo Matching

A Large Dataset to Train Convolutional Networks for Disparity, Optical Flow, and Scene Flow Estimation

Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches

Leveraging stereo matching with learning-based confidence measures

Object scene flow for autonomous vehicles

FlowNet: Learning Optical Flow with Convolutional Networks

Non-parametric Local Transforms for Computing Visual Correspondence

Accurate and Efficient Stereo Processing by Semi-Global Matching and Mutual Information

Field of Study

Journal Information

Name

Page

Venue Information

Name

Type

URL

Alternate Names