Facial Landmark Detection Using Generative Adversarial Network Combined with Autoencoder for Occlusion

Figure 1

The architecture of generative adversarial network combined with improved autoencoders. Our proposed framework mainly consists of three modules: generator (autoencoder), global discriminator, and local discriminator. The generator takes the occluded images as input and outputs the generated images. The two discriminators are used only for training the generator cooperatively while they are not needed in the testing; they are learned to distinguish the generated contents in the occlusion and whole generated images as real and fake.