Some wrongs with the pre-trained model crossformer-b.pth

[hongsheng-Z]

Hi, thanks for your great work. I am using your crossformer_base as my backbone network for downstream tracking tasks. But now when I load your pre-trained model, a very correct Unexpected key(s) appears. My loading code is as follow: ckpt = torch.load(ckpt_path, map_location='cpu') missing_keys, unexpected_keys = backbone.body.load_state_dict(ckpt['model'], strict=False)

The result as follow: unexpected keys: ['norm.weight', 'norm.bias', 'head.weight', 'head.bias', 'layers.0.blocks.0.attn.biases', 'layers.0.blocks.0.attn.relative_position_index', 'layers.0.blocks.1.attn.biases', .....

[cheerss]

It is reasonable that there are some unexpected keys.

Wherein, 'norm.weight', 'norm.bias', 'head.weight', 'head.bias' are weights of the classification head, so the downstream models do not contain these weights. Others, e.g., 'layers.0.blocks.0.attn.biases', are indices for the DPB module, and these indices are non-trainable parameters which can also be computed dynamically without affecting the results. (They are computed dynamically for downstream models here to support dynamic image size)

[hongsheng-Z]

Thanks for your reply, it helped me a lot.