I have created a dataset of 36 classes of sign language. All classes have near about 500 or more images (including manually augmented images). But I am not getting a validation accuracy greater than 2.5. Please help.
