How do you load a model from the mode Zoo (say: https://s3.amazonaws.com/pytorch/models/resnet18-5c106cde.pth)
and how do you forward with this model file?
1 Like
To open that file (which is a collection of weights) you can use params = torch.load('resnet18-5c106cde.pth').
To get a ResNet-18, instead, you should use res18 = torchvision.models.resnet18(pretrained=True).
To forward a random image, you can define a FloatTensor, encapsulate it into a Variable and send it to the network.
x = torch.rand(1, 3, 224, 224)
xVar = torch.autograd.Variable(x)
res18(xVar)
-->
Variable containing:
-0.4374 -0.3994 -0.5249 ... -0.5333 1.4113 0.9452
[torch.FloatTensor of size 1x1000]
You can access the output vector by appending .data to your output Variable.
Let me know if it works.
3 Likes
Nice to see you two again 
2 Likes