KeyError: 'unexpected key "module.conv1_1.weight" in state_dict' about enlightengan HOT 6 CLOSED

I've pushed a new version. You can directly pull the new version to avoid this problem.

from enlightengan.

Fix :

In models/base_model.py replace the load_network with the following:

def load_network(self, network, network_label, epoch_label):
save_filename = '%s_net_%s.pth' % (epoch_label, network_label)
save_path = os.path.join(self.save_dir, save_filename)
state_dict = torch.load(save_path)
new_state_dict = OrderedDict()
for k, v in state_dict.items():
name = k[7:] # remove module.
new_state_dict[name] = v
network.load_state_dict(new_state_dict)

from enlightengan.

您好，在经过上述修改后，
并将 new_state_dict = OrderedDict() 修改为 new_state_dict = collections.OrderedDict() 后，
出现KeyError: 'unexpected key "conv10.bias" in state_dict'，该如何解决？

from enlightengan.

Is your python 3.5, pytorch 0.3.1? And are you trying to run via CPU or GPU?

from enlightengan.

@Dear-Mr @SreeHarshaNelaturu
This problem is about torch.nn.DataParallel(). Normally DataParallel() will cover the original model, which means you can only use model.module to access the original model. And also, if you directly save the model covered by DataParallel, the key of saved model will be module.conv.weight instead of conv.weight.
You can simply use DataParallel to cover the model before you load the pre-trained model to solve this problem.

Eg.

model = ResNet()
parallel_model = torch.nn.DataParallel(model)
print(parallel_model.module)
# Here parallel_model.module is equal to model

from enlightengan.

Is your python 3.5, pytorch 0.3.1? And are you trying to run via CPU or GPU?

Thanks for your reply.
My workspace is python 3.5, PyTorch 0.3.1, torchvision 0.2.0. And i tried to run via GPU.

from enlightengan.