Enhancing Object Detection in Swin Transformer via Global Focusing Mechanism and Data Diversification

The Tiny-ImageN et dataset challenges object detection due to variable object sizes, complex backgrounds, and imbalanced data, which existing Swin Transformer models fail to address adequately. This paper proposes an object detection algorithm that integrates a global focusing mechanism with a data diversification strategy. The global focusing mechanism enhances the model's ability to detect objects of various sizes, while data diversification, through techniques like rotation, clipping, and scaling, improves generalization and reduces overfitting. Experiments on Tiny-ImageNet show the promising performance compared with the state-of-the-arts methods through offering innovative solutions to persistent object detection challenges.