-
您好,在您的项目中应该没有考虑实验可重复性的问题吧,就是即使是相同的代码和数据集,每次重新运行,最后结果都会稍有不同吧。我尝试使实验可重复,因为想公平比较模型的性能,如果本身就有随机性的话,就无法确定是代码本身随机性导致最后结果不同,还是模型的改进了。我在step2_TrainAndValidate.py(主要训练模型的那个文件中)开头添加了如下代码:
![image](https://user…
sjjdd updated
3 years ago
-
where can i download this dataset ?
i need your help.
-
您好,感谢你分享的优秀方案。我在自己数据集上复现时遇到一些问题,想跟您交流下:
数据集:
我使用TNSCUI公开的训练集重新按照8:2划分为训练集(2915)和测试集(729),然后按照你的预处理方法进行处理生成stage1和stage2的数据。
问题:
按照你描述的步骤,采用5折交叉验证,训练完fold_1 后发现stage1网络在验证集上的IoU(0.804),stage2网络单独在验…