训练时准确率几乎接近1,但是预测时准确率不到0.8 #12177
Replies: 6 comments
-
看起来是过拟合 |
Beta Was this translation helpful? Give feedback.
-
应该不是,训练的时候,训练准确率一直比测试准确率高十几个点 |
Beta Was this translation helpful? Give feedback.
-
这个是有点过拟合了,建议加一些数据增强,或者分析下预测错误的是哪些文字,针对性分析下问题 |
Beta Was this translation helpful? Give feedback.
-
请问可以用一些合成数据增强吗? |
Beta Was this translation helpful? Give feedback.
-
请问微调之后识别率有所改善吗?然后您使用的是啥古籍数据集呀 |
Beta Was this translation helpful? Give feedback.
-
我后来换了模型了(捂脸),古籍数据集叫MTHv2,好像是华南理工大学建的 |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
rt,我用了一个古籍数据集训练,模型是ResNet34+Bilstm+CTC,训练的效果很好,但预测起来就是不太行,请问具体是什么原因呢?yml文件的一些超参数如下:
Beta Was this translation helpful? Give feedback.
All reactions