Transformer-Based Image Segmentation: A Comprehensive Overview of Recent Advancements

In 2021, Strudel et al. [31] proposed Segmenter, a network structure consisting entirely of Transformer layers, which utilizes the global image context at each layer of the model. The introduction of this network has greatly enhanced the capabilities of Transformer in the field of image segmentation. In the same year, Chen et al. [32] presented TransUNet, which combines the advantages of Transformer and U-Net for medical image segmentation. Cao et al. [33] introduced Swin-Unet, which is built upon Swin-Transformer blocks and demonstrates excellent segmentation performance and generalization ability in medical image segmentation.