24小时热门版块排行榜

返回列表

【悬赏金币】回答本帖问题，作者smallbug2000将赠送您 20 个金币

smallbug2000

木虫 (著名写手)

应助: 13 (小学生)
金币: 2299.8
散金: 1794
红花: 17
帖子: 1239
在线: 781.9小时
虫号: 434011
注册: 2007-08-18
性别: GG
专业: 计算机应用技术

[求助] 投计算机方面SCI期刊，大修，给点修改意见已有1人参与

投稿到一个计算机方面的SCI期刊，审稿很快，不到2个月给结果，要求大修，下面是我的审稿意见，应该如何修改？大家看下中的概率大不？

Editor's Comments:

1. Please clarify the motivation of this paper.

2. Please give a thorough analysis on the runtime of the proposed method.

3. Please correct all the typos and revise unclear text.

Reviewer Comments:

Reviewer #1 (Reviewer Comments Required):

The authors describe a method for 3D object tracking based on perspective SIFT correspondences. The proposed approach is invariant to viewpoint changing. To show the usefulness of their approach the authors present two comparative evaluations in qualitative as well as quantitative forms.

Although I like the idea described by the authors and I find the work challenging and interesting, I found several points that need to be discussed or improved in order to make the article ready for publication. Moreover, the paper still contains many typos and unclear text passages. Probably, a native English speaker or at least a spell checker could help to improve the paper in terms of presentation quality.

From a content perspective the paper lacks a motivating example, maybe an illustrative figure could be used right from the beginning to make the novel contribution much clearer to the reader.

I am also not confident with the technical descriptions in the paper. Although a lot of math is used and seems to be correct, I am very disappointed about the algorithmic modelling. The authors use lots of formulas in Section 2, but the algorithms fail in this respect. They are written as pseudo code without a lot of technical background, i.e., there are no input parameters nor are there output parameters. Moreover, the data transformations are explained in textual form and should be better described by transforming those parameters. This is the major concern that I have with the current state of the paper. The mathematical issues should definitely be fixed before publication.

Based on those algorithmic concepts I am also asking about runtime complexities which are worth discussing in order to understand if the proposed technique is still interactively usable. From my point of view, it is not clear how fast the approach works and if it can be used in real-time scenarios for example. I wish to see some examples where the approach has limitations, i.e., it would be very important to discuss limitations and scalability issues of the technique in a separate section. This is currently totally unclear which is due to the missing parameter descriptions in the algorithmic concepts.

Figure 3 illustrates a lot of the research but a comparison of the different settings should be made clearer and should be discussed in more detail. Annotations in the figure would also help to understand the changes. Also a legend with value settings is missing in all of the figures. Figure 1 is not aesthetically appealing and should be enhanced by a suitable color coding indicating the different steps and stages.

I am also wondering if the authors could tell a bit more about future work which is rather weak at the moment. The article is written in a way as if many points are still not clear to the authors themselves, hence the missing parameters and also a missing future work. The authors should try to work in this direction and also should try to find some more future scenarios where the approach might be applicable, e.g., in eye tracking research for 3D dynamic stimuli.

In general, the work is interesting and I appreciate the work done by the authors, but in its current form it is still not publishable. Consequently, I am voting for a major revision and hope to see a revised version with a reaction on the reviewers' comments.

Minor comments:

- Longer caption for Figure 1!
- viewpoint- invariant ==> viewpoint-invariant
- interestpoint ==> interest point
- Ref. [X] ==> X et al.[X] describe
- changes of 3D ==> changes of the 3D
- from first several times ==> ???
- iteratively refinement ==> iterative refinement
- for the further ==> for further
- The detail explanation ==> For the detail explanation
- Figure 1: fetaures ==> features
- green arrow ==> a green arrow
- algorithm X ==> Algorithm X
- Where (...) ==> ???
- of L' is ==> of L' are
- equation X ==> Equation X
- we enforce regularity ==> we enforce a regularity
- we used KD-tree ==> we used a KD-tree
- correspondence establish ==> ???
- Caption of Figure 3 too long on page 9!!!
- CPR is the so called Jaccard coefficient! Cardinality of sets is missing in the formula!!!
- video sequences ==> video sequence
- deceases ==> decease, decrease???
- from BU dataset ==> from the BU dataset
- higher accuracy as ==> higher accuracy than
- , but the ==> ???
- on i3 2.0 GHz ==> on an i3 2.0 GHz
- of 3D viewpoint ==> of the 3D viewpoint
- viewpoint -invariance ==> viewpoint-invariant

Reviewer #2 (Reviewer Comments Required):

The work is technically sound and shows an improvement in tracking accuracy as compared to state of the art feature-based tracking methods.
However, I would like to see a more thorough analysis of runtime. The authors rely on a variable number of features combined with iterative methods. This might lead to an unstable per-frame runtime, which contradicts an eventual real-time tracking application.
Also, I would require a discussion on whether this approach can be accelerated by using GPU hardware, which is definitely the case for the competing methods.

Regarding formatting: the experiments section may be too long compared to the first part of the paper. The large number of small images is useless on a printed version of the paper. Fewer and larger images would better demonstrate the tracker capability.

Reviewer #3 (Reviewer Comments Required):

This was a well written paper, with only minor edits and comments. The writing was clear and correct, the methodology appears sound, and the results are well described & presented. Attached are a few specific comments which the authors should review. However, I feel no need to see an updated manuscript: please just correct the few minor grammatical items and consider my other comments which are advisory, not mandatory in nature.

回复此楼