swinv2 backbone Causing map to be 0 during training and inference process

[everytimeis80]

I used the Swin Transformer v2 backbone and Corresponding weights, which resulted in all the training process maps being 0

The weight loading log is as follows _IncompatibleKeys(missing_keys=['norm1.weight', 'norm1.bias', 'norm2.weight', 'norm2.bias', 'norm3.weight', 'norm3.bias'], unexpected_keys=['norm.weight', 'norm.bias', 'layers.0.blocks.1.attn_mask', 'layers.1.blocks.1.attn_mask', 'layers.2.blocks.1.attn_mask', 'layers.2.blocks.3.attn_mask', 'layers.2.blocks.5.attn_mask'])

Has anyone encountered any related issues.I will be deeply grateful for your reply