GIFT: Learning Transformation-Invariant Dense Visual Descriptors via Group CNNs