This repository provides the official implementation of LIIF (Local Implicit Image Function), a CVPR 2021 Oral paper. The method learns continuous image representations that can render images at arbitrary resolutions, enabling arbitrary-scale super-resolution. It uses coordinate-based neural networks to predict RGB values at continuous coordinates, with local feature aggregation from an encoder. Pre-trained models for DIV2K are provided for EDSR and RDN backbones.