self.pos_embed ：Dimension mismatch

[jianjun0407]

The shape of the original input image is (256,256,600). After the previous processing, the shape of the current input x of ImageEncoderViT is (600,3,256,256). After self.patch_embed(), The shape of x is （600,16,16,768）.

However, the shape of self.pos_embed is (1,64,64,768).

so at line 113, x = x + sel.pos_embed cannot be implemented