CBAM Paper Walkthrough: The Double-Attention Mechanism

The Convolutional Block Attention Module (CBAM) is an attention mechanism designed to enhance feature quality in CNN-based models. It consists of two sequential sub-modules: a Channel Attention Module (CAM) and a Spatial Attention Module (SAM). The CAM uses both global max-pooling and average-pooling to learn "what" features are important, while the SAM performs similar pooling across channels to learn "where" to focus in the spatial dimensions. The refined feature map retains the original input dimensions, enabling easy integration into backbone architectures like ResNet, and its design is justified by ablation studies from the original paper.