应《网络安全法》要求，自2017年10月1日起，未进行实名认证将不得使用互联网跟帖服务。为保障您的帐号能够正常使用，请尽快对帐号进行手机号验证，感谢您的理解与支持！

24小时热门版块排行榜    

CyRhmU.jpeg
小木虫论坛-学术科研互动平台 » 专业学科区 » 信息科学 » 信息综合 » 哥的论文大修，哥很郁闷，哥不想改了
191/2返回列表12下一页
查看: 3313  |  回复: 18
只看楼主@他人 存档 新回复提醒 (忽略) 收藏    在APP中查看

smallbug2000

木虫 (著名写手)

[求助] 哥的论文大修，哥很郁闷，哥不想改了

投了一篇论文到计算机视觉相关的杂志，最近返回结果，三个审稿人，其中两个审稿人意见较少。另一个觉得算法的创新性不够，给了大修，悲剧的是这篇论文之前投到CVPR去，也是这个审稿人审的（CVPR据了我），因为审稿意见基本一样，故可以断定是同一审稿人。最终编辑给了大修，但是我很绝望，因为给大修的审稿人从一开始就不认可我的算法，我想既然我认真修改了，可能还是会被他据。将编辑及大修审稿人意见附后，在此向各位大牛咨询两件事：1）该论文是否值得大修，修后被据的概率多大？2）如果大修，如果对付审稿人这么尖锐的问题。在些感谢，对于好的建议发放金币。
Dear**,

Reviewers have now commented on your paper. You will see that they are advising that you revise your manuscript. If you are prepared to undertake the work required, I would be pleased to reconsider my decision.
The reviewers' comments can be found at the end of this email or can be accessed by following the provided link.



Reviewers' comments:

The paper was reviewed by three experts on the topic. The reviewers agree that the paper discusses an important problem worth pursuing. They, however, also unanimously point out short comings of the paper, most important of which is the lack of thorough theoretical and experimental validation of the claims as well as references to prior work. The paper needs major revision to be considered for publication. If the authors decide to submit a major revision, please make sure to address the points raised by the reviewers.

Reviewer #2: : This paper deals with the problem of visual tracking with irregular object motion. The particle set shift approach based on analytic optimization is proposed to deal with the incorrect state dynamic model. Particles are first sampled by the state dynamic model and they are moved to higher likelihood regions by newton optimization by maximizing likelihoods. The efficacy of the proposed approach is demonstrated via experiments with real sequences.

- Positive points: Practically effective approach
- Negative points: Not novel approach, rather heuristic, not convincing experimental results

: The main problem of this paper is the proposed approach is not novel. The proposed approach is quite similar to [18] except the fact that the proposed approach used newton optimization instead of mean shift.

: The proposed approach is rather heuristic. In the algorithm, particle weights are only proportional to likelihood, and this holds true for SIR particle filter where particles are sampled from state dynamic model. However, since particles are moved to higher likelihoods artificially, the weights determined from likelihoods will no longer correctly represent the true posterior. The proposed approach is rather similar to the particle swarm optimization-based tracking, e.g., "" by X. ** et al in CVPR 2008. Another problem of the proposed approach is that it might result in worse tracking results when there are appearance changes caused by various issues such as pose and lighting changes.

: Since the proposed work is based on particle filter, the optimal importance functions used for visual tracking are also relevant. If we can use the optimal importance function, the irregular motion can be handled at least partially. Thus the following papers should be cited and commented as related work:
-

: The supplementary video result is too limited. Only a result for a single short sequence is not sufficient to support the validity of the proposed approach. Why is there no rotational motion in the results? The paper says that the state is translation, scale, and rotation, but there is no rotational motion in the results in the paper and video.

: There must be cases that all particles are outside the basin of convergence. Since the particle are shifted by local optimization, particles will diverge from the optimal positions and tracking will fail. In the proposed approach, there is no consideration of this possibility.

: The optimization will increase the computational complexity considerably. It is necessary to compare the other algorithms under the same computational time, e.g., proposed framework with 100 particles and standard particle filter with 1000 particles.

My current recommendation is major revision. If the authors want to make the paper accepted in spite of the limited novelty and contribution, I think the followings should be addressed in the revision:
- Additional experiments with object appearance changes like illumination
- Addition of a mechanism to deal with the local optima problems with additional experiments with related videos
- Comparison under the same computational complexity for different tracking algorithms

[ Last edited by smallbug2000 on 2012-9-6 at 09:52 ]
回复此楼

» 收录本帖的淘帖专辑推荐

文章写作

» 猜你喜欢

» 本主题相关价值贴推荐，对您同样有帮助:
1楼2012-09-06 09:25:04
已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖
回帖支持 ( 显示支持度最高的前 50 名 )

flywithyu

铜虫 (小有名气)

【答案】应助回帖

★ ★ ★ ★ ★ ★ ★ ★ ★ ★
感谢参与，应助指数 +1
smallbug2000: 金币+2, ★★★很有帮助 2012-09-06 14:45:35
smallbug2000: 金币+2, ★★★很有帮助 2012-09-06 17:37:34
smallbug2000: 金币+2, ★★★很有帮助 2012-09-11 19:27:09
smallbug2000: 金币+4, ★★★★★最佳答案 2012-09-13 14:47:15
这个修改意见不算多的，具体就是：
1、增加实验结果以验证你方法的有效性；（这个只要有数据，应该好办）
2、局部最优问题，这个较难回答，一种你找到方法；另外一种就是通过实验来说明；
3、与不同算法的比较。首先找找网上有没有现成的能出结果的代码吧；实在不行只有自己实现；话说实验室如果有做换个方向的话应该是有一定基础吧。
强烈建议：按照专家意见修改，不要回避，最终你会发现有收获的！！
赞一下(3人) 回复此楼
3楼2012-09-06 12:27:52
 已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖

wangzhipengj

至尊木虫 (文坛精英)
加油 坚持到底才有胜利
赞一下(1人) 回复此楼
4楼2012-09-06 13:05:39
 已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖

nlyang

禁虫 (著名写手)
★ ★
感谢参与，应助指数 +1
smallbug2000: 金币+2, ★★★很有帮助 2012-09-06 17:27:12
本帖内容被屏蔽
6楼2012-09-06 15:55:48
 已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖

flywithyu

铜虫 (小有名气)
引用回帖:
9楼: Originally posted by smallbug2000 at 2012-09-06 18:53:30
好像3 Comparison under the same computational complexity for different tracking algorithms不是只与其它算法的比较吧，应该是比较时间复杂度吧？...

哦，没有仔细看，是的，是比较计算复杂度
我曾经有篇文章，3个审稿人都提了若干条修改意见；其中一个估计是老外，看的非常仔细，提了9条意见，每条都很头疼。
硬着头皮逐条对照修改，实在改不了或者实现起来有难度，也一定要能说服审稿人。
最后改完，3人全部通过。不过等于是重写了篇。。。
还是那句话，审稿人提的意见一般都是很中肯的，建议对照修改，而不是降低档次改投别的。。
赞一下(1人) 回复此楼
10楼2012-09-06 20:54:39
 已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖
普通回帖

carsonwen

铜虫 (小有名气)

【答案】应助回帖

★ ★
感谢参与，应助指数 +1
smallbug2000: 金币+2 2012-09-06 10:35:47
如果不差论文 毕业或评职称的，那就慢慢改吧，挺好的，有专业人士的意见。
当然你也可以折中一点，用他的意见去做一篇新的文章，原来的文章那就再找个地方投了。
赞一下 回复此楼
2楼2012-09-06 10:30:45
 已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖

寒山听雪

铁虫 (小有名气)

【答案】应助回帖

★
感谢参与，应助指数 +1
smallbug2000: 金币+1, ★有帮助 2012-09-07 09:22:26
大修，说明，还是有机会被接受的，你只要按那人要求做就行了。最后论文肯定还是要那审稿人看的，祝好运。
赞一下 回复此楼
R包qkerntool机器学习的最新有效工具大包
5楼2012-09-06 15:42:20
 已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖

wygvern

木虫 (职业作家)

【答案】应助回帖

★ ★
感谢参与，应助指数 +1
smallbug2000: 金币+2, ★★★很有帮助 2012-09-06 17:26:07
呵呵。好好改吧，这个意见真的不算尖锐，我的都是reject and resubmitted照样录用呢，楼主加油吧。越是意见尖锐，录取的可能性越大，但是一定要认真修改。
赞一下 回复此楼
7楼2012-09-06 16:47:13
 已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖

smallbug2000

木虫 (著名写手)
引用回帖:
7楼: Originally posted by wygvern at 2012-09-06 16:47:13
呵呵。好好改吧，这个意见真的不算尖锐，我的都是reject and resubmitted照样录用呢，楼主加油吧。越是意见尖锐，录取的可能性越大，但是一定要认真修改。

很难改的，改不动呀，审稿人太犀利了
赞一下 回复此楼
8楼2012-09-06 17:26:40
 已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖

smallbug2000

木虫 (著名写手)
引用回帖:
3楼: Originally posted by flywithyu at 2012-09-06 12:27:52
这个修改意见不算多的，具体就是：
1、增加实验结果以验证你方法的有效性；（这个只要有数据，应该好办）
2、局部最优问题，这个较难回答，一种你找到方法；另外一种就是通过实验来说明；
3、与不同算法的比较。 ...

好像3 Comparison under the same computational complexity for different tracking algorithms不是只与其它算法的比较吧，应该是比较时间复杂度吧？
赞一下 回复此楼
9楼2012-09-06 18:53:30
 已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖
相关版块跳转 我要订阅楼主 smallbug2000 的主题更新
191/2返回列表12下一页
普通表情 龙 兔 虎 猫
信息提示
关闭
请填处理意见
关闭