AE-GAN-Net: Learning Invariant Feature Descriptor to Match Ground Camera Images and a Large-Scale 3D Image-Based Point Cloud for Outdoor Augmented Reality.