Learning to reconstruct an image from pixel coordinates
Train a neural network to map from x and y coordinate of each pixel to the pixels r,g,b values.
The network has 2 inputs (x and y), several fully connected (highway) layers and three outputs (r, g and b).