The paper "CAMixerSR: Only Details Need More “Attention”" addresses the challenge of large image (2K-8K) super-resolution (SR) by integrating two existing approaches: content-aware routing and token mixer refinement. The authors propose a Content-Aware Mixer (CAMixer) that adaptively assigns convolution for simple contexts and deformable window-attention for complex textures. CAMixer uses a learnable predictor to generate multiple bootstraps, including offsets for window warping, a mask for classification, and convolutional attentions, to modulate attention and improve representation capability. The predictor also introduces a global classification loss to enhance accuracy. CAMixerSR, built on CAMixer, achieves superior performance in lightweight SR, large-input SR, and omnidirectional-image SR, demonstrating a significant improvement over existing methods in terms of quality-complexity trade-offs. The paper includes detailed experimental results and ablation studies to validate the effectiveness of CAMixer and its components.The paper "CAMixerSR: Only Details Need More “Attention”" addresses the challenge of large image (2K-8K) super-resolution (SR) by integrating two existing approaches: content-aware routing and token mixer refinement. The authors propose a Content-Aware Mixer (CAMixer) that adaptively assigns convolution for simple contexts and deformable window-attention for complex textures. CAMixer uses a learnable predictor to generate multiple bootstraps, including offsets for window warping, a mask for classification, and convolutional attentions, to modulate attention and improve representation capability. The predictor also introduces a global classification loss to enhance accuracy. CAMixerSR, built on CAMixer, achieves superior performance in lightweight SR, large-input SR, and omnidirectional-image SR, demonstrating a significant improvement over existing methods in terms of quality-complexity trade-offs. The paper includes detailed experimental results and ablation studies to validate the effectiveness of CAMixer and its components.