WingsNet中注意力模块的疑惑
Closed this issue · 2 comments
Itsanewday commented
haozheng-sjtu commented
3D的中间结果太占显存了,所以原始分辨率的层就用简单的,下采样后可以用更复杂的。
发件人: "Itsanewday" ***@***.***>
收件人: "haozheng-sjtu/3d-airway-segmentation" ***@***.***>
抄送: "Subscribed" ***@***.***>
发送时间: 星期五, 2023年 2 月 10日 上午 11:37:37
主题: [haozheng-sjtu/3d-airway-segmentation] WingsNet中注意力模块的疑惑 (Issue #11)
作者您好,首先感谢公开支气管分割框架,对于我的帮助很大!看了您的《Alleviating Class-wise Gradient Imbalance for
Pulmonary Airway Segmentation》后,在网络设计方面有一些疑惑,您参考的论文将SE拓展为cSE(原始的), sSE(WingsNet采用的)和scSE(二者结合)的,参考的论文中做过一些对照实验,结果表明第三类变种在医学分割任务中性能最好,您在WingsNet中为什么采用sSE而不是scSE呢?
—
Reply to this email directly, [ #11 | view it on GitHub ] , or [ https://github.com/notifications/unsubscribe-auth/ATQRWJLHBD4LOAWEJJCDYSTWWWZYDANCNFSM6AAAAAAUXKGXNA | unsubscribe ] .
You are receiving this because you are subscribed to this thread. Message ID: <haozheng-sjtu/3d-airway-segmentation/issues/11 @ github . com>
Itsanewday commented
确实,谢谢
…---原始邮件---
发件人: ***@***.***>
发送时间: 2023年2月16日(周四) 晚上8:21
收件人: ***@***.***>;
抄送: ***@***.******@***.***>;
主题: Re: [haozheng-sjtu/3d-airway-segmentation] WingsNet中注意力模块的疑惑 (Issue #11)
3D的中间结果太占显存了,所以原始分辨率的层就用简单的,下采样后可以用更复杂的。
发件人: "Itsanewday" ***@***.***>
收件人: "haozheng-sjtu/3d-airway-segmentation" ***@***.***>
抄送: "Subscribed" ***@***.***>
发送时间: 星期五, 2023年 2 月 10日 上午 11:37:37
主题: [haozheng-sjtu/3d-airway-segmentation] WingsNet中注意力模块的疑惑 (Issue #11)
作者您好,首先感谢公开支气管分割框架,对于我的帮助很大!看了您的《Alleviating Class-wise Gradient Imbalance for
Pulmonary Airway Segmentation》后,在网络设计方面有一些疑惑,您参考的论文将SE拓展为cSE(原始的), sSE(WingsNet采用的)和scSE(二者结合)的,参考的论文中做过一些对照实验,结果表明第三类变种在医学分割任务中性能最好,您在WingsNet中为什么采用sSE而不是scSE呢?
—
Reply to this email directly, [ #11 | view it on GitHub ] , or [ https://github.com/notifications/unsubscribe-auth/ATQRWJLHBD4LOAWEJJCDYSTWWWZYDANCNFSM6AAAAAAUXKGXNA | unsubscribe ] .
You are receiving this because you are subscribed to this thread. Message ID: <haozheng-sjtu/3d-airway-segmentation/issues/11 @ github . com>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: ***@***.***>