search

Xiyue-Wang / RetCCL

GNU General Public License v3.0
96 stars 8 forks source link

直接使用该特征提取器时patch像素大小的问题 #6

Open LiuJin629 opened 11 months ago

LiuJin629 commented 11 months ago

你好，标准的ResNet输入图像要求是224×224，但是我切的patch大小是512×512，请问我如果直接使用你训练好的特征提取器提特征的话，可以直接将这些patch投入到模型中吗，还是说需要resize一下，如果可以直接投入，那对下游任务的预测效果会有影响吗

LiuJin629 commented 11 months ago

在您的RetCCL这盘论文图像的预处理中，我发现对不同中心的WSI切成的patches大小还不一样(有的是1024×1024，有的是224×224)，请问最后在您训练最终的特征提取器模型时，对这些不同大小的patches大小都做了resize吗，resize成多少了呢

Xiyue-Wang commented 10 months ago

Keeping the size at 5x (224x224) or 10x (224x224) is fine, we've tested it and it works well, as long as all the data stays on the same scale.

LiuJin629 commented 10 months ago

thank you！

---- 回复的原邮件 ---- | 发件人 | @.> | | 发送日期 | 2023年12月09日 01:18 | | 收件人 | Xiyue-Wang/RetCCL @.> | | 抄送人 | LiuJin @.>, Author @.> | | 主题 | Re: [Xiyue-Wang/RetCCL] 直接使用该特征提取器时patch像素大小的问题 (Issue #6) |

Keeping the size at 5x (224x224) or 10x (224x224) is fine, we've tested it and it works well, as long as all the data stays on the same scale.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>