Experiments with super resolution.

Currently modified versions of VDSR and LapSRN for 2x are implemented, trained model on game sprites is on models folders, train parameters are in experiments folder.

Papers based on:

• Deep Laplacian Pyramid Networks for Fast and Accurate Super-Resolution
• Accurate Image Super-Resolution Using Very Deep Convolutional Networks
• Online Learning Rate Adaptation with Hypergradient Descent
• Stochastic Gradient Descent with Warm Restarts

Other references:

• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• Deconvolution and Checkerboard Artifacts