Efficient Video Super-Resolution via Two-Step 2D SR and Image-to-Image Conditional Generation

Video super-resolution (VSR) has gained significant attention due to its potential to enhance video quality. However, existing VSR models are often computationally heavy, limiting their practical application. To address this, we propose a two-step approach: applying 2D super-resolution (SR) followed by image-to-image conditional generation. Initial experiments using Mean Squared Error (MSE) show significant improvements. Replacing MSE with Mean Absolute Error (MAE) and blending it with Structural Similarity Index Measure (SSIM) further enhances performance. Our method achieves superior results with reduced computational complexity, offering a practical solution for real-world VSR applications. Our code is available at https://github.com/msdsm/efficient-video-sr-2steps.