search

microsoft / CSWin-Transformer

CSWin Transformer: A General Vision Transformer Backbone with Cross-Shaped, CVPR 2022
MIT License
539 stars 78 forks source link

The weight size of pretrained model Cswin-L is not the same with the script? #34

Open Sunting78 opened 2 years ago

Sunting78 commented 2 years ago

[04/20 19:11:50] d2.checkpoint.c2_model_loading WARNING: merge3.conv.weight will not be loaded. Please double check and see if this is desired. [04/20 19:11:50] d2.checkpoint.c2_model_loading WARNING: Shape of merge3.norm.bias in checkpoint is torch.Size([1152]), while shape of backbone.merge3.norm.bias in model is torch.Size([576]). [04/20 19:11:50] d2.checkpoint.c2_model_loading WARNING: merge3.norm.bias will not be loaded. Please double check and see if this is desired. [04/20 19:11:50] d2.checkpoint.c2_model_loading WARNING: Shape of merge3.norm.weight in checkpoint is torch.Size([1152]), while shape of backbone.merge3.norm.weight in model is torch.Size([576]). [04/20 19:11:50] d2.checkpoint.c2_model_loading WARNING: merge3.norm.weight will not be loaded. Please double check and see if this is desired. [04/20 19:11:50] d2.checkpoint.c2_model_loading WARNING: Shape of stage4.0.attns.0.get_v.bias in checkpoint is torch.Size([1152]), while shape of backbone.stage4.0.attns.0.get_v.bias in model is torch.Size([576]). [04/20 19:11:50] d2.checkpoint.c2_model_loading WARNING: stage4.0.attns.0.get_v.bias will not be loaded. Please double check and see if this is desired. [04/20 19:11:50] d2.checkpoint.c2_model_loading WARNING: Shape of stage4.0.attns.0.get_v.weight in checkpoint is torch.Size([1152, 1, 3, 3]), while shape of backbone.stage4.0.attns.0.get_v.weight in model is torch.Size([576, 1, 3, 3]).

Why the the weights in Cswin-L can not totally loaded?

ChristinaXu4work commented 1 year ago

I got the same problem. Have you solved that？