Transformer-based Segmentation: Addressing Challenges in Railway Image Analysis

These networks utilize the Transformer architecture to effectively capture contextual information within images, leading to impressive segmentation results. However, it is important to note that these segmentation networks have predominantly been applied to medical images and natural scene images in previous studies. In the case of railway image datasets, the available data is limited and the images themselves are complex. Consequently, several existing networks may not be well-suited for accurate railway image segmentation.