Image Completion with Incomplete Data

Start Image Completion with Incomplete Data Most of the image completion and generative models require fully-observed samples to train the network. However, as it’s stated in the ambientGAN paper, obtaining high resolution samples can be very expensive or impractical for some applications. The model in this project combines ideas from the following works: AmbientGAN Globally and Locally Consistent Image Completion The AmbientGAN model enables training a generative model directly from noisy or incomplete samples. The...