Error(s) in loading state_dict for DenseNet:

vision
1

Hello,

I would like to load the weights of the densenet121 network like that :

net = models.densenet121(pretrained=False)
net.load_state_dict(torch.load(’./densenet121-a639ec97.pth’))

it return me this error :
RuntimeError: Error(s) in loading state_dict for DenseNet:
Missing key(s) in state_dict: “features.denseblock1.denselayer1.norm1.weight”, “features.denseblock1.denselayer1.norm1.bias”, “features.denseblock1.denselayer1.norm1.running_mean”, “features.denseblock1.denselayer1.norm1.running_var”, “features.denseblock1.denselayer1.conv1.weight”, “features.denseblock1.denselayer1.norm2.weight”, “features.denseblock1.denselayer1.norm2.bias”, “features.denseblock1.denselayer1.norm2.running_mean”, “features.denseblock1.denselayer1.norm2.running_var”, “features.denseblock1.denselayer1.conv2.weight”
etc…
…

I use this method instead of :
net = models.densenet121(pretrained=True)
because the proxy of my company don’t allow to do it…

Could someone help me ?

Thank you

2

I used to load model’s weights like this:

net = models.densenet121(pretrained=False).cuda()
net.load_state_dict(torch.load('path/to/saved_model.pth.tar')['state_dict'])

1 Like
3

Thank you, you answer guided me to look at why my network don’t know state dict so is decide to use the constructor code from pytorch densenet and it works :
import re

Code modified from torchvision densenet source for loading from pre .4 densenet weights.

checkpoint = torch.load(’./model.pth.tar’)
state_dict = checkpoint[‘state_dict’]
remove_data_parallel = False # Change if you don’t want to use nn.DataParallel(model)

pattern = re.compile(
r’^(.*denselayer\d+.(?:norm|relu|conv)).((?:[12]).(?:weight|bias|running_mean|running_var))$’)
for key in list(state_dict.keys()):
match = pattern.match(key)
new_key = match.group(1) + match.group(2) if match else key
new_key = new_key[7:] if remove_data_parallel else new_key
state_dict[new_key] = state_dict[key]
# Delete old key only if modified.
if match or remove_data_parallel:
del state_dict[key]

1 Like