The paper introduces the Convolutional Block Attention Module (CBAM), a lightweight and effective attention mechanism for feed-forward convolutional neural networks (CNNs). CBAM sequentially infers attention maps along two dimensions: channel and spatial, and multiplies these maps to the input feature map for adaptive feature refinement. The module is designed to be seamlessly integrated into any CNN architecture with minimal overhead and can be end-to-end trained. Extensive experiments on ImageNet-1K, MS COCO, and VOC 2007 datasets demonstrate consistent improvements in classification and detection performances, validating the wide applicability and effectiveness of CBAM. The code and models are publicly available. Key contributions include the proposal of CBAM, extensive ablation studies, and performance improvements across multiple benchmarks.The paper introduces the Convolutional Block Attention Module (CBAM), a lightweight and effective attention mechanism for feed-forward convolutional neural networks (CNNs). CBAM sequentially infers attention maps along two dimensions: channel and spatial, and multiplies these maps to the input feature map for adaptive feature refinement. The module is designed to be seamlessly integrated into any CNN architecture with minimal overhead and can be end-to-end trained. Extensive experiments on ImageNet-1K, MS COCO, and VOC 2007 datasets demonstrate consistent improvements in classification and detection performances, validating the wide applicability and effectiveness of CBAM. The code and models are publicly available. Key contributions include the proposal of CBAM, extensive ablation studies, and performance improvements across multiple benchmarks.