Segmentation accuracy and network architecture in the code.

[mumu0310]

I ran into a few issues while running the code and I hope to get some clarifications. 1. I ran the code 150K times on the ISIC dataset, but the Dice score of the sampling results is only 0.77, which is a bit off from what's in the paper. 2. While reading through the code, I noticed that some structures in the network are not present, including the SS-former and the processing of anchor embeddings.