Issue in Loading State Dict for FairDetector: Missing Keys

[zzy1130]

Hi, thank you very much for your work, but I still got some problems. When I directly started from the testing part, I encountered an error while trying to load the state dictionary for the FairDetector model in the test.py script. The partial error message is as follows:

RuntimeError: Error(s) in loading state_dict for FairDetector:
        Missing key(s) in state_dict: "encoder_f.last_linear.weight", "encoder_f.last_linear.bias", "encoder_f.adjust_channel.0.weight", "encoder_f.adjust_channel.0.bias", "encoder_f.adjust_channel.1.weight", "encoder_f.adjust_channel.1.bias", "encoder_f.adjust_channel.1.running_mean", "encoder_f.adjust_channel.1.running_var", "encoder_c.conv1.weight", "encoder_c.bn1.weight", "encoder_c.bn1.bias", "encoder_c.bn1.running_mean", "encoder_c.bn1.running_var", "encoder_c.conv2.weight", "encoder_c.bn2.weight", "encoder_c.bn2.bias", "encoder_c.bn2.running_mean", "encoder_c.bn2.running_var", "encoder_c.block1.skip.weight", "encoder_c.block1.skipbn.weight", "encoder_c.block1.skipbn.bias", "encoder_c.block1.skipbn.running_mean", "encoder_c.block1.skipbn.running_var", "encoder_c.block1.rep.0.conv1.weight", "encoder_c.block1.rep.0.pointwise.weight", "encoder_c.block1.rep.1.weight", "encoder_c.block1.rep.1.bias", "encoder_c.block1.rep.1.running_mean", "encoder_c.block1.rep.1.running_var", "encoder_c.block1.rep.3.conv1.weight", "encoder_c.block1.rep.3.pointwise.weight", "encoder_c.block1.rep.4.weight", "encoder_c.block1.rep.4.bias", "encoder_c.block1.rep.4.running_mean", "encoder_c.block1.rep.4.running_var", "encoder_c.block2.skip.weight", "encoder_c.block2.skipbn.weight", "encoder_c.block2.skipbn.bias", "encoder_c.block2.skipbn.running_mean"
...

for missing keys in the model, and

 Unexpected key(s) in state_dict: "encoder_f.fc.weight", "encoder_f.fc.bias". 
        size mismatch for encoder_f.block1.rep.0.pointwise.weight: copying a param with shape torch.Size([128, 64]) from checkpoint, the shape in current model is torch.Size([128, 64, 1, 1]).
        size mismatch for encoder_f.block1.rep.3.pointwise.weight: copying a param with shape torch.Size([128, 128]) from checkpoint, the shape in current model is torch.Size([128, 128, 1, 1]).
        size mismatch for encoder_f.block2.rep.1.pointwise.weight: copying a param with shape torch.Size([256, 128]) from checkpoint, the shape in current model is torch.Size([256, 128, 1, 1]).
        size mismatch for encoder_f.block2.rep.4.pointwise.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([256, 256, 1, 1]).
        size mismatch for encoder_f.block3.rep.1.pointwise.weight: copying a param with shape torch.Size([728, 256]) from checkpoint, the shape in current model is torch.Size([728, 256, 1, 1]).
        size mismatch for encoder_f.block3.rep.4.pointwise.weight: copying a param with shape torch.Size([728, 728]) from checkpoint, the shape in current model is torch.Size([728, 728, 1, 1]).
        size mismatch for encoder_f.block4.rep.1.pointwise.weight: copying a param with shape torch.Size([728, 728]) from checkpoint, the shape in current model is torch.Size([728, 728, 1, 1]).
        size mismatch for encoder_f.block4.rep.4.pointwise.weight: copying a param with shape torch.Size([728, 728]) from checkpoint, the shape in current model is torch.Size([728, 728, 1, 1]).
        size mismatch for encoder_f.block4.rep.7.pointwise.weight: copying a param with shape torch.Size([728, 728]) from checkpoint, the shape in current model is torch.Size([728, 728, 1, 1]).
        size mismatch for encoder_f.block5.rep.1.pointwise.weight: copying a param with shape torch.Size([728, 728]) from checkpoint, the shape in current model is torch.Size([728, 728, 1, 1]).
        size mismatch for encoder_f.block5.rep.4.pointwise.weight: copying a param with shape torch.Size([728, 728]) from checkpoint, the shape in current model is torch.Size([728, 728, 1, 1]).
        size mismatch for encoder_f.block5.rep.7.pointwise.weight: copying a param with shape torch.Size([728, 728]) from checkpoint, the shape in current model is torch.Size([728, 728, 1, 1]).
        size mismatch for encoder_f.block6.rep.1.pointwise.weight: copying a param with shape torch.Size([728, 728]) from checkpoint, the shape in current model is torch.Size([728, 728, 1, 1]).
        size mismatch for encoder_f.block6.rep.4.pointwise.weight: copying a param with shape torch.Size([728, 728]) from checkpoint, the shape in current model is torch.Size([728, 728, 1, 1]).
        size mismatch for encoder_f.block6.rep.7.pointwise.weight: copying a param with shape torch.Size([728, 728]) from checkpoint, the shape in current model is torch.Size([728, 728, 1, 1]).
        size mismatch for encoder_f.block7.rep.1.pointwise.weight: copying a param with shape torch.Size([728, 728]) from checkpoint, the shape in current model is torch.Size([728, 728, 1, 1]).
        size mismatch for encoder_f.block7.rep.4.pointwise.weight: copying a param with shape torch.Size([728, 728]) from checkpoint, the shape in current model is torch.Size([728, 728, 1, 1]).
...

for unexpected keys and checkpoint mismatch from the checkpoint.

It seems like there is a mismatch between the checkpoint you provided and the model structure. Can you suggest any solutions to this problem? Or is it originated from some other mistakes that I made? Thank you for your attention to this matter.

[Purdue-M2]

Hi Thank you for your interest in our work. The checkpoint you loaded is for the backbone only, not the entire model, which is why you encountered such an issue.