Next Article in Journal
Structural Design of Ocean Temperature and Depth Sensor with Quick Response and High Sensitivity
Next Article in Special Issue
Five-Direction Occlusion Filling with Five Layer Parallel Two-Stage Pipeline for Stereo Matching with Sub-Pixel Disparity Map Estimation
Previous Article in Journal
Trusted Multi-Domain DDoS Detection Based on Federated Learning
Previous Article in Special Issue
A Reconfigurable Visual–Inertial Odometry Accelerated Core with High Area and Energy Efficiency for Autonomous Mobile Robots
 
 
Article
Peer-Review Record

The mr-MDA: An Invariant to Shifting, Scaling, and Rotating Variance for 3D Object Recognition Using Diffractive Deep Neural Network

Sensors 2022, 22(20), 7754; https://doi.org/10.3390/s22207754
by Liang Zhou 1,2, Jiashuo Shi 1,2 and Xinyu Zhang 1,2,*
Reviewer 1:
Reviewer 2:
Sensors 2022, 22(20), 7754; https://doi.org/10.3390/s22207754
Submission received: 5 September 2022 / Revised: 3 October 2022 / Accepted: 10 October 2022 / Published: 12 October 2022
(This article belongs to the Special Issue Computer Vision and Sensor Technology)

Round 1

Reviewer 1 Report

Comments to the Author

In this work, based on the multiple-view D2NN array (MDA), 3D objects can be recognized in a high-speed and dynamic way. Through numerical verification, the author finds that it is invariant to typical shifting, scaling and rotating variance of the target in complicated situation. This is a very interesting work. The paper is well organized and clearly written, which can be accepted for publication, provided that the authors further improve the work by addressing the comments as follows.

Minor comments,

1. The network proposed by the author is very complex, but the accuracy is still very low (no more than 90%) when shifting, scaling and rotating variance for 3D objects are not considered. Whether the author can analyze the causes of low accuracy and the solutions.

2. For the optical neural network with diffractive residual blocks, what is its diffraction efficiency theoretically?

3. In Figure 9, it seems that the neural network proposed by the author has no obvious improvement in shifting, scaling and rotating object recognition compared with the previous MDA. What is the main reason and how to improve it?

4. What are the advantages of the neural network proposed by the author over previous methods for 2D object recognition?

 

5. In the introduction, the author mentioned the extensive application of electronic neural networks (ENNs) and the diffractive deep neural network (D2NN). The description is too simple. More previous works are suggested to be mentioned (Gu, Min, et al. "Optically digitalized holography: a perspective for all-optical machine learning." Engineering 5.3 (2019): 363-365. Luan, Haitao, et al. "768-ary Laguerre-Gaussian-mode shift keying free-space optical communication based on convolutional neural networks." Optics Express 29.13 (2021): 19807-19818.) to be helpful to wide readers.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

lines 44:48 please explain what architecture you mean. Do that statement include ReLU activation function?  How regularization (BathNorm or other) can be included into this area.

lines 52:59 add link,

all equations: please, add conventional denotation for all variables you use.

section 4: i do not find any explanation of non-linearity (activation function) beside eq 11 (last layer).

please, do not use such long sentences, it leads to complexity of reading. 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Back to TopTop