Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

VIG中使用的KNN dilation #237

Open
yuan0038 opened this issue Nov 14, 2023 · 8 comments
Open

VIG中使用的KNN dilation #237

yuan0038 opened this issue Nov 14, 2023 · 8 comments

Comments

@yuan0038
Copy link

大佬,想请教一下在VIG中不同层中KNN设置不同的dilation是出于什么动机呢,为什么不直接全设置为1?

@iamhankai
Copy link
Member

主要是为了让图卷积能够观测到更大范围的近邻节点,提升聚合的多样性

@yuan0038
Copy link
Author

懂了,谢谢大佬!因为觉得您的这个工作很猛,所以想在您的这个工作基础上做探索。还想问您一个问题,就是您觉得单纯用图像和标签训练,有没有可能让模型学习到语义,就比如说让鸟头这个patch的K近邻都为鸟的部分,树枝的K近邻都是树枝。。
51_156_pad_split_out

@iamhankai
Copy link
Member

有可能啊,把dilation参数调小点就更集中了

@yuan0038
Copy link
Author

要得~,第一次见能回issues的大佬😂

@yuan0038
Copy link
Author

yuan0038 commented Jan 5, 2024

大佬,想问一下① 下图中的our ViG 具体是哪个ViG ,diversity是在哪个数据集上测的(ImageNet的话,是验证集吗) ② diversity 和 dilated rate 之间的关系 image

@yuan0038
Copy link
Author

yuan0038 commented Jan 5, 2024

韩凯大佬,我是科研细🐶,还想请教您一个困扰我很久的问题,就是我往模型中添加新模块后,如何分析模块生效与否背后的原因?(比如 有一个CNN,我往其中加了一个新模块,掉点掉的很多,有没有什么办法来了解这个模块为什么不work,从而做到有针对性地改进模块,而不是盲目地替换模块和调参。)

@iamhankai
Copy link
Member

不敢称大佬。。你可以:1)可视化:分析一下新模块的输出特征图,可视化一下,和之前的特征图对比一下看看,比较直观;2)定量分析:比如用vig论文中的diversity评价指标测测特征图的diversity变化情况。等等

@yuan0038
Copy link
Author

yuan0038 commented Jan 8, 2024

大佬谦虚了。看论文刷到您好多次了😂,还在github上帮人指点迷津,我直接当您迷弟,You are my hero🌹🌹🌹!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants