I am using MatConvNet for the classification two different pants.
i am using MINST example architecture of 28x28 input image.
the dataset I have for two classes are ~14000 images after data augmentation. 45% for training and testing and 10% for testing.
While i was getting overfitting.
The Original architecture is like that:
Conv1-relu1-pool1 Conv2-relu2-pool2 Conv3-relu3 Conv4-relu4
To avoid overfitting I have added dropout layers.
Conv1-relu1-pool1 dropout 70, Conv2-relu2-pool2 dropout 70, Conv3-relu3 dropout 50, Conv4-relu4 dropout 70
But still its showing overfitting at the output..
Any suggestions please to solve this issue
Asked
Active
Viewed 437 times
2

Addee
- 151
- 4
-
This might help: http://stackoverflow.com/questions/7509666/when-should-i-stop-training-neural-network-for-classification-using-cross-valida – Arun Jose May 11 '17 at 06:57
-
thanks but how this is related? it is how to stop training. however, in my case I am not even getting right training. – Addee May 11 '17 at 07:07
-
If you have found a solution to your problem, please considering answering your question. Otherwise, please update your question with more detailed info about your model, including learning rate, etc. – NULL Aug 01 '17 at 11:57
-
`the dataset I have for two classes are ~14000 images after data augmentation. 45% for training and testing and 10% for testing.` Please modify your question and fix the percentages. – NULL Aug 01 '17 at 11:58
-
See: https://stats.stackexchange.com/questions/365778/what-should-i-do-when-my-neural-network-doesnt-generalize-well – Sycorax Sep 13 '18 at 22:25