Test after training leads to 0.0 dice

[pervinh]

I trained my model for 2000 epochs. Loss value decreased with every batch. However when i use test.py with the same training data, i obtain 0.0 dice.

mean dice for class-1 is 0.0 mean dice for class-2 is 0.0

What could be the reason for this?

[Dylanself123]

i found the pre-processing step may cause this problem

[pervinh]

@Dylanself123 is it the normalization step in brain18.py?

[Yufengevan]

Hi,Have you resolve this problem? I have the same problem.

[Dylanself123]

@Dylanself123 is it the normalization step in brain18.py?

The crop step, leading the input images different between training and test. U can check the training_data_process and testing_data_process

[infinite-tao]

Did you solve the problem? I have the same problem here, very confused.

[nuist-xinyu]

@Dylanself123 is it the normalization step in brain18.py?

The crop step, leading the input images different between training and test. U can check the training_data_process and testing_data_process

Did you solve the problem? I have the same problem here, very confused.

[infinite-tao]

Thanks! I tried skipping the data processing,the training process is normal, loss is always falling, but the test is very unstable (sometimes dice = 0.2, sometimes dice = 0). Maybe I have another problem here.

Well happy every day!

On 11/26/2019 14:07，nuist-xinyu[email protected] wrote：

@Dylanself123 is it the normalization step in brain18.py?

The crop step, leading the input images different between training and test. U can check the training_data_process and testing_data_process

Did you solve the problem? I have the same problem here, very confused.

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

[nuist-xinyu]

Thanks! I tried skipping the data processing,the training process is normal, loss is always falling, but the test is very unstable (sometimes dice = 0.2, sometimes dice = 0). Maybe I have another problem here. Well happy every day! On 11/26/2019 14:07，nuist-xinyu[email protected] wrote： @Dylanself123 is it the normalization step in brain18.py? The crop step, leading the input images different between training and test. U can check the training_data_process and testing_data_process Did you solve the problem? I have the same problem here, very confused. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

Thanks! I tried skipping the data processing,the training process is normal, loss is always falling, but the test is very unstable (sometimes dice = 0.2, sometimes dice = 0). Maybe I have another problem here. Well happy every day! On 11/26/2019 14:07，nuist-xinyu[email protected] wrote： @Dylanself123 is it the normalization step in brain18.py? The crop step, leading the input images different between training and test. U can check the training_data_process and testing_data_process Did you solve the problem? I have the same problem here, very confused. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

Hello, thank you for your reply. I have recently used 3D segmentation to use this model, There are two types of goals I segmented，but when I tested, there was only one label and no other. How can I solve this problem, this problem has troubled me for a long time

Well happy every day!

[wangm-ting]

Did you solve the problem? I have the same problem here, very confused.

when I use my own dataset as the test ,the dice=0 .Do you kown how to solve it？

[lander1003]

Did you solve the problem? I have the same problem here, very confused.

when I use my own dataset as the test ,the dice=0 .Do you kown how to solve it？

Do you solve it now?

[lander1003]

Thanks! I tried skipping the data processing,the training process is normal, loss is always falling, but the test is very unstable (sometimes dice = 0.2, sometimes dice = 0). Maybe I have another problem here. Well happy every day! On 11/26/2019 14:07，nuist-xinyu[email protected] wrote： @Dylanself123 is it the normalization step in brain18.py? The crop step, leading the input images different between training and test. U can check the training_data_process and testing_data_process Did you solve the problem? I have the same problem here, very confused. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

Did you solve it ?