Reconstructing seen image from brain activity by visually-guided cognitive representation and adversarial learning.