error using existing state dict in the checkpoint folder

[HeidiHe]

trafficstars

I was trying to install this model on my computer. I noticed that the program by default uses the existing state dict in the checkpoint folder, but I got the following error: """ Missing key(s) in state_dict: "conv1.weight", "conv1.bias", "conv2.weight", "conv2.bias", "conv3.weight", "conv3.bias", "conv4.weight", "conv4.bias", "conv5.weight", "conv5.bias", "conv1_s.weight", "conv1_s.bias", "conv2_s.weight", "conv2_s.bias", "conv3_s.weight", "conv3_s.bias", "conv4_s.weight", "conv4_s.bias", "transConv1.weight", "transConv1.bias", "transConv2.weight", "transConv2.bias", "conv6.weight", "conv6.bias", "skipConv1.0.weight", "skipConv1.0.bias", "skipConv1.2.weight", "skipConv1.2.bias", "skipConv1.4.weight", "skipConv1.4.bias", "skipConv2.weight", "skipConv2.bias", "skipConv3.weight", "skipConv3.bias". Unexpected key(s) in state_dict: "module.conv1.weight", "module.conv1.bias", "module.conv2.weight", "module.conv2.bias", "module.conv3.weight", "module.conv3.bias", "module.conv4.weight", "module.conv4.bias", "module.conv5.weight", "module.conv5.bias", "module.conv1_s.weight", "module.conv1_s.bias", "module.conv2_s.weight", "module.conv2_s.bias", "module.conv3_s.weight", "module.conv3_s.bias", "module.conv4_s.weight", "module.conv4_s.bias", "module.transConv1.weight", "module.transConv1.bias", "module.transConv2.weight", "module.transConv2.bias", "module.conv6.weight", "module.conv6.bias", "module.skipConv1.0.weight", "module.skipConv1.0.bias", "module.skipConv1.2.weight", "module.skipConv1.2.bias", "module.skipConv1.4.weight", "module.skipConv1.4.bias", "module.skipConv2.weight", "module.skipConv2.bias", "module.skipConv3.weight", "module.skipConv3.bias". """ I wonder if there is a ready-to-use pretrained model for testing? Or I have to train my own model. I encountered this great paper and want to test it out myself. However I have limited computational power on my current computer (very unfortunately a MAC computer, no CUDA available), it'd be very helpful if there is a pretrained model avaliable for testing.

Thank you!

[BingyaoHuang]

The existing state dict in the checkpoint folder was trained on GPU and it needs an extra step if you want to load it to a CPU model. Please see this tutorial for more details.