test error in resnet.py

[yja1]

python script/experiment/train_pcb.py \ -d '(0,)' \ --only_test true \ --dataset duke \ --exp_dir /data/exp_directory \ --model_weight_file /data/pcb_model_weights/duke/model_weight.pth

the error is: duke test set

NO. Images: 31969 NO. IDs: 751 NO. Query Images: 3368 NO. Gallery Images: 15913 NO. Multi-query Images: 12688

Traceback (most recent call last): File "script/experiment/train_pcb.py", line 495, in main() File "script/experiment/train_pcb.py", line 304, in main train_set = create_dataset(**cfg.train_set_kwargs) File "./bpm/model/PCBModel.py", line 19, in init self.base = resnet50(pretrained=True, last_conv_stride=last_conv_stride) File "./bpm/model/resnet.py", line 190, in resnet50 model.load_state_dict(remove_fc(model_zoo.load_url(model_urls['resnet50']))) File "./bpm/model/resnet.py", line 152, in remove_fc for key, value in state_dict.items(): RuntimeError: OrderedDict mutated during iteration

[yja1]

I use your pcb_model_weights

[ecr23xx]

Maybe you should use python 2 and I modify the remove_fc function in resnet.py:

def remove_fc(state_dict):
  """Remove the fc layer parameters from state_dict."""
  no_fc_dict = state_dict.copy()
  for key, value in state_dict.items():
    if key.startswith('fc.'):
      del no_fc_dict[key]
return no_fc_dict

[yja1]

right! thx

[huanghoujing]

@yja1 @ECer23 Thank you very much for pointing out this dictionary discrepancy between python 2 and 3.

[ghostbody]

comprehensions is better:

def remove_fc(state_dict):
    return {key: value for key, value in state_dict.items() if key.startswith('fc.')}

[huanghoujing]

@ghostbody Thanks for your good idea! Just one thing, here you miss the not. The correct version is

def remove_fc(state_dict):
    return {key: value for key, value in state_dict.items() if not key.startswith('fc.')}

[ghostbody]

@huanghoujing yes, u r right.