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

关于先验位置关系 #29

Open
mitu752 opened this issue Nov 18, 2024 · 1 comment
Open

关于先验位置关系 #29

mitu752 opened this issue Nov 18, 2024 · 1 comment
Labels
question Further information is requested

Comments

@mitu752
Copy link

mitu752 commented Nov 18, 2024

Question

作者,你好,很棒的工作。想问下作者有没有尝试过通过计算标签(groundtruth)的位置关系得到先验位置关系设计位置编码器呢。

补充信息

No response

@mitu752 mitu752 added the question Further information is requested label Nov 18, 2024
@xiuqhou
Copy link
Owner

xiuqhou commented Nov 19, 2024

我并没有尝试过这种思路,感觉预先统计的先验可能更倾向于反应“平均”后的关系,但不能灵活地反应样本间的方差。
但确实有这么做的论文,比如在我related work中有提到,有的方法会从Visual Genome数据库中挑选出所需要的类别来统计各种先验关系,然后固定下来作为一个全局的特征或网络参数,这个数据库有很多relation标注。小样本学习中也有类似的做法,他们会用预训练的网络(比如ResNet)提取每个类别的平均特征,然后把这些特征固定下来作为网络参数,用来和后续的输入样本做交互。
但是总体来说,我觉得这个思路更适用于同类别样本方差较小或者样本较少的情况,这样预先提取的先验才足够代表大部分样本,比如长尾分布或者小样本任务。

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

No branches or pull requests

2 participants