plutoyuxie / AutoEncoder-SSIM-for-unsupervised-anomaly-detection-

Improving Unsupervised Defect Segmentation by Applying Structural Similarity to Autoencoders
121 stars 25 forks source link

你好,请问运用这个数据集和代码重建的图很模糊,和image里的效果相差甚远,是什么原因呢?谢谢。 #4

Closed SunnysideBao closed 3 years ago

plutoyuxie commented 4 years ago

您好,image里的grid重建结果,我的设置是 epoch 200次; 数据增强augment_num: 10000; z_dim: 100; 希望能对您有帮助。

SunnysideBao commented 4 years ago

谢谢您的回复,训练的数据集您用的是grid里面good的数据吗?要求输入大小是128*128,您是将它分块还是等比例缩小呢?

plutoyuxie commented 4 years ago

你好,是的,用无缺陷数据训练。 先缩放再随机裁剪,数据增强有旋转翻转等。

发自我的iPhone

------------------ 原始邮件 ------------------ 发件人: Sunnyside Bao <notifications@github.com> 发送时间: 2020年7月12日 09:22 抄送: plutoyuxie <346929320@qq.com>, Comment <comment@noreply.github.com> 主题: 回复:[plutoyuxie/AutoEncoder-SSIM-for-unsupervised-anomaly-detection-] 你好,请问运用这个数据集和代码重建的图很模糊,和image里的效果相差甚远,是什么原因呢?谢谢。 (#4)

谢谢您的回复,训练的数据集您用的是grid里面good的数据吗?要求输入大小是128*128,您是将它分块还是等比例缩小呢?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

SunnysideBao commented 4 years ago

谢谢,原本的像素值为10241024,您是缩放到什么程度进行裁剪到128128呢?

plutoyuxie commented 4 years ago

256x256

发自我的iPhone

------------------ 原始邮件 ------------------ 发件人: Sunnyside Bao <notifications@github.com> 发送时间: 2020年7月12日 09:40 抄送: plutoyuxie <346929320@qq.com>, Comment <comment@noreply.github.com> 主题: 回复:[plutoyuxie/AutoEncoder-SSIM-for-unsupervised-anomaly-detection-] 你好,请问运用这个数据集和代码重建的图很模糊,和image里的效果相差甚远,是什么原因呢?谢谢。 (#4)

谢谢,原本的像素值为10241024,您是缩放到什么程度进行裁剪到128128呢?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.