densenet/Alexnet finetuning

vision
1

image
image807×216 10.5 KB

This is the last layers of densenet from the pytorch hub models. I added an extra linear layer after the given linear layer in the following way.

import torch.nn as nn
model = torch.hub.load('pytorch/vision:v0.6.0', 'densenet121', pretrained=True)
modules = list(model.children()) 
new = nn.Sequential(*modules,
                    nn.Linear(model.classifier.out_features, 100))

But when I use

x = torch.ones((1,3, 224, 224))

out = new(x)

out

I get the followin error-

image
image947×350 16.8 KB

Why so? Aren’t the dimensions right here?
Similar things happen when adding a new layer to Alexnet too. but it works if I edit the dimension of the existing layers.

2

Hi,

The problem comes with the way of constructing the new model. Please use the following code snippet. It will work.

import torch.nn as nn
import torch

model = torch.hub.load('pytorch/vision:v0.6.0', 'densenet121', pretrained=True)
n_model = nn.Sequential(model,
                    nn.Linear(model.classifier.out_features, 100))
x = torch.ones((1,3, 224, 224))
out = n_model(x)
print(out.shape)

Thanks

1 Like