Skip to main content
. 2021 Nov 25;11:22908. doi: 10.1038/s41598-021-02095-4

Figure 4.

Figure 4

The proposed SAE block, which consists of three blocks: a NL context modeling block, a SE transforming block and a residual block. It takes template features and search region features as inputs, and calculates the spatial attentions of both branches.