Next Article in Journal
A New GaN HEMT Small-Signal Model Considering Source via Effects for 5G Millimeter-Wave Power Amplifier Design
Previous Article in Journal
Development of Intellectual Web System for Morph Analyzing of Uzbek Words
Previous Article in Special Issue
Automated Surface Defect Inspection Based on Autoencoders and Fully Convolutional Neural Networks
 
 
Article
Peer-Review Record

EAR-Net: Efficient Atrous Residual Network for Semantic Segmentation of Street Scenes Based on Deep Learning

Appl. Sci. 2021, 11(19), 9119; https://doi.org/10.3390/app11199119
by Seokyong Shin 1, Sanghun Lee 2,* and Hyunho Han 3
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Appl. Sci. 2021, 11(19), 9119; https://doi.org/10.3390/app11199119
Submission received: 20 August 2021 / Revised: 27 September 2021 / Accepted: 28 September 2021 / Published: 30 September 2021
(This article belongs to the Special Issue Computing and Artificial Intelligence for Visual Data Analysis)

Round 1

Reviewer 1 Report

The authors propose the atrous residual network for semantic segmentation of street scenes. Generally speaking, the manuscript is well organized. The authors perform numerical simulations on real datasets and compare their model with different models. The result indicates that the proposed model may outperform.  However, there is one place for which the authors need to further clarify:

1) In table 4 of 4.3.2 section (DSConv analysis), the authors mention the EAR-Net with standard convolution. The authors need to further clarify the term "standard convolution" since readers may have a misunderstanding between traditional convolution (figure 2 (a)) and atrous convolution( figure 2 (b)).  The authors need to point out which convolution they are using.

2) The authors need to add a numerical experiment in table 4, since they need to compare their models by using three different convolutions, i.e. traditional convolution, atrous convolution, and depthwise separable convolution (DSConv). This will further verify that their model is better. The added simulation depends on what the standard convolution is. If the standard convolution is atrous convolution,  the authors need to add the simulation by using the traditional convolution in their network. But if the standard convolution is traditional convolution,  the authors need to add the simulation by using the atrous convolution in their network. 

Author Response

Please see the attachments.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper proposes an effective ear net for street scene semantic segmentation, which can improve the accuracy while maintaining the computational cost. The method has certain practical significance. However, there are some suggestions that should be considered in this article:

(1) For the analysis of DSConv, it is recommended to conduct efficiency comparative analysis for specific hardware.

(2) It is suggested that this method is compared with the best algorithm for semantic segmentation of cityscapes dataset.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Back to TopTop