How to return weights of all convolution layers in a VGG16 model?

vision
1

Hi,
I have a vgg16 model with features below: And I want to retrieve all weight values of the conv layers
VGG16Model (
(features): Sequential (
(0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU (inplace)
(2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU (inplace)
(4): MaxPool2d (size=(2, 2), stride=(2, 2), dilation=(1, 1))
(5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(6): ReLU (inplace)
(7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(8): ReLU (inplace)
(9): MaxPool2d (size=(2, 2), stride=(2, 2), dilation=(1, 1))
(10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(11): ReLU (inplace)
(12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(13): ReLU (inplace)
(14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(15): ReLU (inplace)
(16): MaxPool2d (size=(2, 2), stride=(2, 2), dilation=(1, 1))
(17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(18): ReLU (inplace)
(19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(20): ReLU (inplace)
(21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(22): ReLU (inplace)
(23): MaxPool2d (size=(2, 2), stride=(2, 2), dilation=(1, 1))
(24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(25): ReLU (inplace)
(26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(27): ReLU (inplace)
(28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(29): ReLU (inplace)
(30): MaxPool2d (size=(2, 2), stride=(2, 2), dilation=(1, 1))
)

What I am doing is this:

weights_to_be_used = [
p for n, p in net.features._modules.items()
if isinstance(p, torch.nn.modules.conv.Conv2d )]

This is what is returned:
[Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)), Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))]

But I am actually trying to get the values of the weights in all the conv layers. Any help will be appreciated. Thanks

2

p is of torch.nn.Parameter() type, which is a thin wrapper over underlining torch Tensor. p.data will give you the actual tensor.

3

Hi @jdhao , thanks for the reply. I’m not sure i understand entirely because I’m still quite new to pytorch. Is there an example code snippet?

4

Just a little modification of your original code should be enough,

weight_to_be_used = [p.data for n, p in net.features._modules.items() if isinstance(p, nn.modules.conv.Conv2d)]
5

somehow i got an error that p has no attribute data but changing p.data to p.weight seems to work. Thank you

6

Yes, my fault. The following should work,

weight_to_be_used = []
for mod in net.features.children():
    if isinstance(mod, nn.modules.conv.Conv2d)
    weight_to_be_used.append(mod.weight.data)

mod is all the modules in net.features, convolution modules has two parameters: weight and bias, which can be accessed by mod.weight and mod.bias attribute. The underlying tensor can be accessed using data attribute.

1 Like
7

I see. Learned something from that. Thanks