润色：In the task of semantic segmentation the encoder needs to remap the semantic information learned by the encoder to the pixel space through the decoder and finally achieve pixel-intensive prediction

In the task of semantic segmentation, the decoder plays a crucial role in remapping the semantic information learned by the encoder back to the pixel space, ultimately achieving pixel-level prediction. However, the downsampling and other operations performed by the encoder result in a decrease in the resolution of the feature image. This leads to the loss of edge information and a portion of the image context, resulting in a significant loss of detailed information in the features used for restoration.

To address this issue, several strategies are employed. Firstly, the encoder leverages both CNN and Transformer modules to extract more comprehensive global information. This combination allows for a richer representation of the input image, capturing both local and global context.

Additionally, a jump connection between the encoder and the decoder is established. This connection enables the features from both the encoder and the decoder to be connected, ensuring that the high-level features are involved in the decoding process. By incorporating these high-level features, the information loss of the encoder is further reduced, leading to improved segmentation results by the decoder.

Overall, these techniques help to mitigate the information loss caused by downsampling and enhance the segmentation performance by leveraging both global and local information