Extract Features from models made with nn.ModuleList

vision
1

Hi;

I am using a 103 layer Tiramisu as mentioned here https://github.com/bfortuner/pytorch_tiramisu.
Now I would like to extract features from the bottleneck layer of the trained network, The authors have made use of nn.ModuleList for costructing the network. Any suggestions .

2

Rewriting forward function would be the easiest.
Hook would also works.

3

Okay thanks @checyuntc, I modified the forward function. Now, I would like to take the network use layers upto the bottleneck and add an additional class

4

Hi;

I tried tweaking the forward pass and I think it was the easiest thing to do. However is there an nice way to truncate the model till the bottleneck layer, because I would now like to add an average pooling at the bottleneck layer to perform regression. Any ideas

5

here is an example of extract features from vgg with nn.Sequential. nn.ModuleList should be similar.

from torch import nn
from torchvision.models import vgg16

class Vgg16(nn.Module):
    def __init__(self):
        super(Vgg16, self).__init__()
        features = list(vgg16(pretrained = True).features)[:23]
        # the output of 3,8,15,22 layer is : relu1_2,relu2_2,relu3_3,relu4_3
        self.features = nn.ModuleList(features)
        
    def forward(self, x):
        results = []
        for ii,model in enumerate(self.features):
            x = model(x)
            if ii in {3,8,15,22}:
                results.append(x)
        
        return results
6

Maybe this recipe is useful:

Best regards

Thomas