Skips math part, focuses on why are researchers solve the particular problem, what are the results and so on.
1) Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network
-
Motivation: Finding an efficient way of upscaling from low resolution to high resolution.
-
Proposition: Fractional
$\frac{1}{r}$ convolution is implemented as Pixel Shuffle. PS rearranges elements of a tensor from HxWxC*r^2 to rHxrWxC. -
Results: Better PSNR (dB) - MSE like metric for image resolution, less parameters, faster training and inference.