Compression-Efficient Feature Extraction Method for a CMOS Image Sensor

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have developed a compression efficient feature extraction method for a CMOS image sensor which would senses the target and extract binary feature and there by saving power by operating it in dual modes such as sensing mode and viewing mode.  I would say this paper is well organized and provided enough evidences for their proposed methods by performing some studies. However there are few concerns that needs to be addressed:

1. The authors used lightweight DNN in their proposed model, but did not provide information regarding the network model configuration used and their learning parameters applied to it. So, they need to justify and add relevant information even though they used their model. 
2. Why they limited to one Visual Wake Words (VWW) dataset itself as we have multiple datasets such as "Wake Vision" which has advanced features than the VWW dataset. Wake Vision provides more features with clean data. 
3. The authors need to provide a discussion section where they could address the key challenges they faced in developing the method. It is also advised to add a table to provide the comparison with the other feature extraction models which were developed with CMOS image sensors so far. For instance, there are few methods which focused on the wide dynamic range feature extraction but they have some limitations. So the authors can discuss such relevant methods in this section. 
4. The authors need to update the abbreviation table since many short words were used not explained in the abbreviation table. For instance FLOPs- Floating Point Operations per Second, which were not provided in the whole manuscript. It is advised to check the whole manuscript to update the abbreviation table. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

1. The authors have proposed a compression-efficient feature extraction method for a CMOS image sensor.
2. I think the proposed technique is novel and highly interesting.
3. However, the proposed technique is so theoretical and ambiguous, thus the technique must be proven by experimental results.
4. Figure 2 is a block diagram containing the most important ideas proposed in this paper. However, in my opinion, it is almost impossible to purchase commercial CIS chips that enable the proposed method. How can you implement the technique? Please let me know that what kind of CIS chips are available to realize your technique. I think the authors must design and fabricate a CIS chip that enables the implementation of the proposed idea..
5. Unless the proposed ideas are objectively verified through actual chip fabrication and measurement, the ideas in this paper are merely simulations and thus unlikely to be accepted by a journal. In particular, the results in Figures 5, 6, and 10 are just simple simulation results obtained by the authors using their own computers, making it difficult to ensure objectivity in the paper. Further, since the simulation results are so perfect, I cannot confirm them.
6. Therefore, please resubmit your paper after completing the design, fabrication, and measurement of the relevant chip. For reference, one related paper is introduced below. The paper below has designed, fabricated, and measured a fully custom CIS to implement edge detection, thereby validating the author's idea. Due to the policy of this journal, it is difficult to accept papers based solely on simulation results.

“Suhyeon Lee, Yu Chan Yun, Seung Min Heu, Kyu Hyun Lee, Seung Joon Lee, Kyungmin Lee, Jiin Moon, Hyuna Lim, Taeun Jang, Minkyu Song and Soo Youn Kim, “The Design of a Computer Vision Sensor Based on a Low-Power Edge Detection Circuit”, Sensors, Vol.25, pp.3219-3232, 20th, May, 2025.”

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

According to the reviewer's comments, it has been well revised.