VISOR-NET: Visibility Estimation Based on Deep Ordinal Relative Learning under Discrete-Level Labels

Round 1

Reviewer 1 Report

The article concerns proposes a novel end-to-end pipeline that uses Ordinal information and Relative relation of the images for VISibility estimation.

The authors in the article presented the regression model to estimate visibility under discrete level labels. There is an element of novelty in this approach.

The developed solution was compared with deep learning algorithms.

In my opinion, the article is very good.

Author Response

Comments: The article concerns proposes a novel end-to-end pipeline that uses Ordinal information and Relative relation of the images for VISibility estimation.The authors in the article presented the regression model to estimate visibility under discrete level labels. There is an element of novelty in this approach.The developed solution was compared with deep learning algorithms.In my opinion, the article is very good.

Response: Thank you for your encouraging comments!

Reviewer 2 Report

The paper presents a novel end-to-end pipeline VISOR-NET for visibility estimation with ordinal relative learning. Also, a large-scale dataset FHVI is collected. The visibility label of each image is annotated and manually checked regarding a professional visibility meter. Experiments on the dataset are performed to justify the efficacy and efficiency of the proposed method.

1.     The structure of the Feature Extraction Regression Module is not clear. It is not clear either how the FCRN can realize regression.

2.     The reviewer is not sure what the P and G are in Fig.2. If P means predictions and G ground truth, the relation judgment may be a regression module, but it was not indicated in this work.

3.     From Fig.1, the reviewer can hardly interpret the contribution of the work.

4.     The proposed solution may have some issues for general applications.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf