Final Project for Machine Learning course

In our modern world, we come across a variety of images daily. The multitude of diversity involved in these images is huge. Yet, when we come across an image with a missing or obscured region that we have not seen before, we are still able to imagine its exact content. The way we picture its content is by analysing the image to find the surrounding regions and complete the image coherently. But the most important thing that helps us in making sense of the context is the fact that images are highly structured. This structure helps in extrapolating the recognizable contents of the image to complete the image. This is the motivation behind our project to apply machine learning algorithms to train our system on the image content so that similar to humans they can learn the context of the image. Then use this context to make predictions for the missing parts of the image that come close to the actual image. We use a CNN based Generative Adversarial Network trained using reconstruction and adversarial loss for the filling of missing regions.