Why is the model with high accuracy on the validation set lower on the test set? Is it because the input size of the test set and the validation set are different?

[Arvin117]