Few shot learning (Or one shot learning) for the image classification problem can be used when there are few samples per class in the dataset (One method is siamese networks). Few shot semantic segmentation attempts to do the same but it generates a semantic image label rather than a classification result (See here and here).

So if I have an imbalanced semantic segmentation problem, where some classes have far lesser training data than others (minority classes), would I get any gain by training a few-shot semantic segmentation model to better recognize the minority classes (Given that there is so little data from them)?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.