Difference between Keras model.evaluate() and cross-validation scores

Question

since I have a really small dataset, I used cross-validation to compute a training/validation error in the following way:

for train, val in cv.split(x_train,