Further experiments in software engineering, machine learning and computer graphics.
Training a CNN for ×2 image super-resolution, then exploring residual learning, learning rates and input degradation.