Long-Range Non-Line-of-Sight Imaging Based on Projected Images from Multiple Light Fields
Round 1
Reviewer 1 Report
1. line 198 a period appears at the beginning of the next line.
2. line 205 and line 218 Whether the meaning of A(I) in Equation 6,7 is the same？
3. line 271-276 for lines that contain symbols throughout the text, optimize line spacing.
4. line327 ，what is 11, add some describes. Is it the first scanning at Position 1?
5. In the light transaction equations, the light position and others parameters is the important factors. But in the simulation , the content is only discussed as H, F, etc. Will the light position and others parameters affect the results?
6. The diffuse wall is complex, and the imaging is very different. Some have strong reflection, and some are not obvious. Will these affect the final recognition result?
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
In this paper, the authors proposed a Non-Line-of-Sight (NLOS) imaging metho with a encoder-decoder network. Results show minor improvements over existing method. I have the following concerns.
1. In section 2.2.1, the detailed mathematical model of the imaging model is presented. However, the model used in section 2.2.2 is a U-net, which is not relevant to the theory. In general, a network is usually designed based on the theory, but the two parts in this paper are quite independent.
2. Only one method is used for performance comparison, which I believe is not convincing enough. More SOTA methods should be added.
3. The experiment is quite simple, and more scenes should be used.
4. As to the application, it seems that the experiment is quite simple, and how can it be used in autonomous driving?
5. The language should be improved. For example, in line 121, 'more close' should be replaced by 'closer'
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
For whether active or passive non-line-of-sight imaging (NLOS), existing methods always have to trade off among real time data acquisition, low set-up cost, long imaging range, high-fidelity hidden object reconstruction, low reconstruction computational complexity and robustness to environmental disturbance.
In this paper, authors proposed an active NLOS scheme which can achieve real time data acquisition, low set-up cost and long imaging range. It is based on a former work proposed by Chen et al. [23]. Compared to the former work, authors added a structural similarity indicator in the deep learning model training to improve structure information reconstruction. Simulation and experiment results demonstrated that the method proposed by authors achieved higher reconstruction fidelity.
I recommend it to be published on Photonics if the following minor questions can be addressed.
1. Compared to Chen’s method, authors only used green laser and add a narrow-band filter before camera. Could authors explain why they did not adopt three lasers like Chen? Also, I noticed that authors used telescopic lens before camera, which was not used in Chen’s setup. I wonder if this brought some advantages, considering this telescopic lens makes the whole imaging setup bulky.
2. Does this method only work for the planar object? Did authors test it on the real 3D hidden object?
3. Since authors emphasize this is for long range imaging, could they provide the distance limit for their method to still work? In the manuscript, they mentioned the distance between the equipment and the NLOS scene was 53.9m in their experiment. Could it be longer? When does the proposed method totally fail?
4. The language can be further polished up.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
I think the revised manuscript have responded to my comments. so I suggest the manuscript should be accept and published.
Reviewer 2 Report
I believe the authors have done a great revision job. My concerns have been addressed and I suggest the paper be accepted.
