This paper presents a novel deep learning architecture, the Cascading Residual Network (CARN), designed for single-image super-resolution (SISR) tasks. The authors address the challenge of achieving high performance while maintaining lightweight and efficient models suitable for real-world applications. CARN introduces a cascading mechanism within a residual network framework, allowing for the integration of multi-level representations and multiple shortcut connections. This mechanism enhances the model's ability to capture and propagate information across different layers, improving both the quality and efficiency of the super-resolution process. The paper also introduces a variant of CARN, called CARN-Mobile (CARN-M), which further optimizes the model for speed and computational efficiency. CARN-M combines an efficient residual block with a recursive network architecture, reducing the number of parameters and operations while maintaining competitive performance. Experiments on various benchmark datasets demonstrate that CARN and CARN-M achieve comparable or superior results to state-of-the-art methods with significantly fewer parameters and operations. The authors also conduct ablation studies to analyze the effectiveness of the cascading modules and the trade-offs between performance, parameters, and operations. The results show that both local and global cascading modules contribute to the model's performance, and the efficient residual block and recursive network scheme significantly enhance computational efficiency. The paper concludes by discussing the potential applications of CARN and CARN-M in video streaming and other real-world scenarios, emphasizing the importance of lightweight and efficient models for practical deployment.This paper presents a novel deep learning architecture, the Cascading Residual Network (CARN), designed for single-image super-resolution (SISR) tasks. The authors address the challenge of achieving high performance while maintaining lightweight and efficient models suitable for real-world applications. CARN introduces a cascading mechanism within a residual network framework, allowing for the integration of multi-level representations and multiple shortcut connections. This mechanism enhances the model's ability to capture and propagate information across different layers, improving both the quality and efficiency of the super-resolution process. The paper also introduces a variant of CARN, called CARN-Mobile (CARN-M), which further optimizes the model for speed and computational efficiency. CARN-M combines an efficient residual block with a recursive network architecture, reducing the number of parameters and operations while maintaining competitive performance. Experiments on various benchmark datasets demonstrate that CARN and CARN-M achieve comparable or superior results to state-of-the-art methods with significantly fewer parameters and operations. The authors also conduct ablation studies to analyze the effectiveness of the cascading modules and the trade-offs between performance, parameters, and operations. The results show that both local and global cascading modules contribute to the model's performance, and the efficient residual block and recursive network scheme significantly enhance computational efficiency. The paper concludes by discussing the potential applications of CARN and CARN-M in video streaming and other real-world scenarios, emphasizing the importance of lightweight and efficient models for practical deployment.