| 查看: 966 | 回复: 17 | |||
| 当前只显示满足指定条件的回帖，点击这里查看本话题的所有回帖 | |||
[交流]
大家看看重投还有希望吗 小白第一次投稿 投的access
|
|||
|
Reviewer: 1 Recommendation: Reject (updates required before resubmission) Comments: The authors proposed a face super-resolution (FSR) network based on a gradient information compensation module. This paper believes that the gradient information is beneficial to the performance of FSR, and designed a special module to extract the gradient information of the feature maps. I think this article is interesting and looks good, but there are still some issues that need to be revised (unordered): (1) In order to prove the effectiveness of the proposed method, the authors conducted a series of ablation studies. However, ablation research must follow some principles. For example, when authors prove that their GRBs are beneficial to improve the performance of the network, they should not prove by reducing the number of GRBs. Because it is not clear whether the performance loss is due to the reduction in the number of parameters or the lack of gradient information. Considering that the gradient layer is an artificial design without any learnable parameters, and its computational complexity is negligible. So it seems more reasonable to delete the gradient layer directly and observe the network performance. (2) The author declares "Since only the test code of FSRNet was released, we re-implement FSRNet with Pytorch". I would suggest highlighting this difference in Table 1 with ‘*’ or similar symbols. (3) When comparing the proposed method with the SOTA method, the computational complexity and parameter amount of each method should also be listed, which will help to study the efficiency of the network. (4) I think the author missed some existing methods, especially the SOTA method since 2020. I think the author needs to add several recent SOTA methods and compare the proposed method. Comments: This manuscript should be resubmitted after being revised according to the follows as Comment 1：The overall framework of Figure 1 needs to be properly described. And some symbols in the figure are confused. Comment 2：RCAB is used in the proposed method, but its introduction and implementation details are missing. Comment 3：In page 6, the GFNet is compared with state-of-the-art methods and achieves better results. However, the comparison methods are put forward from 2014 to 2019. References are not the recent one and some appropriate key studies are missing. Comment 4：In Abstract, “the proposed network is able to reconstruct fine face images with a performance improvement of 0.3 dB PSNR and 0.82 dB PNSR”. This is really confusing. What is the comparison method? How do you get the results of 0.3 dB PSNR and 0.82 dB PNSR? These are not mentioned in the text. Comment 5：The format of references is incorrect, such as [1], [12], [14], [15], [20], etc. Comment 6：To have an unbiased view in the paper, there should be some discussions on the limitations of the method. Comment 7：The grammar needs to be carefully revised. There are many spelling mistakes, such as "dsuper-resolution" on page 2. |
回复此楼» 猜你喜欢
等盲审
已经有5人回复
-
今年面上预计又打酱油了
已经有17人回复
-
七级岗，到手不到8000，带本科论文12教学分/生，额定教学分280
已经有20人回复
-
说同学对待毕业论文不认真后她很不悦
已经有15人回复
-
去美国的华人导师底下读博后会很难出站吗？
已经有4人回复
-
想请问下IEEE Trans最后一轮，你们等了多久？
已经有4人回复
-
考研日记
已经有4人回复
-
博士申请
已经有11人回复
-
求博导收留
已经有4人回复
-
博士毕业找工作选择！
已经有9人回复
» 抢金币啦！回帖就可以得到:
-
坐标成都诚征女友
+1/134
-
南京征一个不介意身高的女友
+1/82
-
燕山大学张利鹏教授招收车辆工程专业学术博士1名
+1/77
-
招募：全职java工程师
+1/73
-
结婚纪念日投稿，期待好运
+1/57
-
坐标郑州，诚征男友
+1/39
-
博士后生活补贴可以算入个人工资，那科研资助也算工资吗？
+1/37
-
南京林业大学 国家优青课题组 招收博士研究生
+1/30
-
寻找一女友
+1/30
-
双非数学图论硕士，求博导！
+1/24
-
UTSW Wang 教授招聘博后，研究科学家，博士生和访问学生
+1/20
-
浙大研究助理招聘
+1/9
-
新加坡南洋理工大学物理系招博士后 （二维材料光谱与器件）
+1/9
-
招2024级南方科技大学-鹏城实验室联培博士生招生-光纤通信的通感一体化、光电传感器件
+1/7
-
有机硅低温固化问题
+1/6
-
中科院苏州生物医学工程技术研究所生物电课题组招聘博士后/特别研究助理
+1/6
-
天津医科大学(211)药理学系招2024年秋季入学申请考核博士研究生(生命科学背景)
+1/5
-
南京林业大学 国家优青课题组 招收博士研究生
+1/3
-
上海交大菜鸟青椒课题组招聘合成化学方向博士后（可转半导体）
+1/2
-
恋爱脑男找恋爱脑女
+1/1
+1/134
5楼2021-11-30 22:53:56
