ResNet50 has gained prominence due to its residual learning architecture, which facilitates the efficient training of very deep models. However, a key limitation of ResNet50 lies in its uniform processing of spatial and channel-wise features, potentially leading to the neglect of critical information. To overcome this shortcoming, this work explores the integration of the Convolutional Block Attention Module (CBAM) into the ResNet50 architecture. The CBAM modules enhance the network’s ability to selectively focus on informative regions and feature channels, thereby improving representational power. Experimental results confirm that the CBAM-augmented ResNet50 achieves significantly improved performance. However, directly inserting new attention modules into a pretrained architecture may degrade performance, as the new components are not immediately adapted to the learned representations. Therefore, fine-tuning the network during training becomes crucial to ensure that the added modules are properly integrated and contribute effectively to the model’s learning process.