Efficient Contextual Feature Network for Single Image Super Resolution

The field of efficient super-resolution techniques has witnessed significant progress, with advancements in reducing parameters and FLOPs and enhancing feature utilization through complex layer connections. However, these methods may not be suitable for resource-constrained devices due to their computational demands. We propose a novel approach called Efficient Contextual Feature Network (ECFN) to address this issue. ECFN utilizes two convolutional layers to learn residual contextual local features, striking a balance between model effectiveness, inference speed, and efficiency. These updates improve performance compared to previously reported efficient super-resolution models for Single Image Super-Resolution (SISR), offering faster runtime without compromising high PSNR or SSIM.