Classes Issue in EMNIST Dataset

Viewed 282

So I'm using the dataset from here https://www.kaggle.com/crawford/emnist namely emnist-letters-train-labels-idx1-ubyte.gz

When I go through it returns in 27 classes whilst there should only be 26 as it's for alphabets. I dug in as much as I can but failed to find in labels for these.

Any help could be great.

1 Answers

A guess from my side would be they left one spare class if the sign is not a letter from the given alphabet or e.g. a number. So if the sign cannot be classified, the 27th spare class ("not a letter") will be assigned to it. Did you already try to feed it some sign which is not a letter and see if it lands in the same class most of the time?

Related