Waifu2x's Potential

Waifu2x is a popular open source software that makes use of super resolution to improve 2D image quality. Not only does this software provide image upscaling, but also provides noise reduction/removal. Unlike traditional upscalers, Waifu2x's upscaler not only increases an image's resolution, but by utilizing deep convolutional neural networks, its able to with suprising accuracy, correctly predict missing pixels. An example of Waifu2x in action can be seen in the image below which can also be found on Waifu2x's github repository. The first section shows how good Waifu2x's 2x upscale of the small image look compared to GIMP's lanczos3 algorithm. The next section shows how Waifu2x performs against GIMP's selective blur when working with a noisy image, and in this case, with compression artifacts. GIMP's selective blur didn't do too bad, but Waifu2x's noise reduction is simply at another level! This image doesn't show the original image before being compressed, resulting in these artifacts, but based on the previous image we can assume that Waifu2x did a pretty solid job. In the final section we have a small noisy image that's being upscaled and denoised all at once. Again you can see how well Waifu2x did, but you can truly see how great the combination of upscaling and denoising if you compare the final result here to the initial 2x upscaling result. By denoising the image, althought it removed a lot of the noise, the denoising function unexpected(or expectectly) performed it across the entire image and caused blurring by smoothing out a lot of things. Most aren't noticable, but the most noticable would be how the black strokes/lines are a lot smoother or "sharper" compared to just upscaling the image(which can be seen in the first image result). So despite how great Waifu2x is at improving an image quality, it is not the holy grail. We will be exploring other derivatives of Waifu2x such as Waifu2x-Caffe and Video2x.

Simple Infographic used by Waifu2x