after adding the information of View Dependence, the training results gets more realistic. However, by increasing model's dependence on view directions, although the model could memorize the traning images better and get lower traning loss, its generalization on unseen viewing directions gets worse by overfitting to the trained directions.