Given partial observations of a natural image, can a network be trained that can accurately predict the missing portion to a realistic standard?


by : 
Kevin F.


This research project will attempt to use mathematical methods as well as convolutional neural networks to predict the content of an image given context surrounding it. Applications include image editing as well as a real-time sensor, allowing the image to have a predicted fill rather than simply a black box. There is also significant research that implies that accomplishing this task may allow for an opportunity to train a network that can successfully represent natural looking images.