This is enough samples.
What is the confusion matrix for training and validation: is the “missing” class equally diluted in the other ones? Or confused with a single other class?
Do you notice the same for training and validation evaluations?
What classifier do you use?
It could be great to have partial logs to help you better.
I find strange that one class is completely missing: In the case of a very hard-to-separate class, I would rather expect a great amount of confusion between classes, but not expecting a class that is completely missing…