MATLAB: How to train net with high class imbalance

artificial intelligencecnndeep learningimage segmentation

Hi, I am doing a semantic segmentation that has a large difference between the two classes. So that, it tends to overpredict when I train the network. Is there any way to train the network specifically in one of the classes?

I am using classweights already, but as most of the pictures is background, the training accuracy tends to increase and the loss to decrease until constant, which generates zero accuracy when testing in the most important but imbalanced class.

Best Answer

Over-sampling can be one solution for that task. Copy the image of smaller classes so that the number of images in each class can balance and perform training using data augmentation function.

https://jp.mathworks.com/matlabcentral/fileexchange/78020-oversampling-for-deep-learning-classification-example

Related Solutions

MATLAB: Dealing imbalanced data in neural network

There many ways to deal with unbalanced classes when there is no more real data available. Over the decades I have used the following

 1. Use the summary statistics of small classes to simulate more data
 2. Design multiple nets using the smaller classes and subsets of the larger classes. 
    Then combine the answers.
 3. Use a cost matrix to enhance the influence of the small subsets 
    and/or reduce the influence of the larger subsets
 4. A combination of the above.

The basis of the techniques can be understood by examining the following term in the Bayesian Risk

 Cij * Pi * p(i|x)

which involves the probability density, a prori probability and the classification cost.

Hope this helps.

Thank you for formally accepting my answer

Greg

MATLAB: Data augmentation_CNN

As @Mohammad Sami already mentioned, evaluating the performance of the neural network based on training accuracy/loss only is not a good idea.

If the accuracy you mentioned is the training accuracy then it may happened that the network simply memorized the training data, rather than learning general features that enable the network to make accurate predictions for new data. To check if your network is overfitting, compare the training loss and accuracy to the corresponding validation metrics. If the training loss is significantly lower than the validation loss, or the training accuracy is significantly higher than the validation accuracy, then your network is overfitting.

Refer to Deep Learning Tips and Tricks, Set Up Parameters and Train Convolutional Neural Network & Preprocess Images for Deep Learning.

Best Answer

Related Solutions

MATLAB: Dealing imbalanced data in neural network

MATLAB: Data augmentation_CNN

Related Question