The Rise of Transformers in Image Segmentation: From GANs to Specialized Architectures

In 2016, Luc et al. [28] used adversarial networks to correct the masks generated by segmentation networks and improve segmentation performance. In 2019, Majurski et al. [29] utilized GANs to achieve cell contour segmentation. Since the birth of Transformers in 2017, they have been rapidly applied in the field of natural language processing (NLP). Subsequently, researchers discovered the great potential of Transformers in computer vision, leading to the proposal of segmentation networks based on Transformers [30]. In 2021, Strudel et al. [31] introduced Segmenter, Chen et al. [32] proposed TransUNet, and Liu et al. [33] presented Swin-Unet. These networks all leverage Transformers to extract image information and have achieved outstanding segmentation results.