Super-resolution is a computer vision task of increasing resolution of the images. The demand for image enhancing comes from mobile devices such as cameras and mobile phones. These devices are computationally restricted, therefore computer vision models which solve super-resolution task must be significantly compressed and optimized before the deployment. Binary neural networks gain popularity nowadays as binarization is a good way to reduce latency and power consumption. The main disadvantage of this approach is that binary models often suffer from significant quality degradation.

The goal of the challenge is to achieve the partially or fully binary models for super-resolution task for x2 and x4 scales. The less complexity metric the model has, the better. The more peak signal-to-noise ratio (PSNR) the model has on test dataset, the better.