I want to use the pre-trained model vgg19
in torchvision.models.vgg
to extract features of ground truth and estimate results from the conv1_1
,conv2_1
,conv3_1
,pool1
,pool2
. Just like the perceptual loss in the neural style transfer.
I have trouble in using pre-trained model to get the feature maps.
You can use the example of fast-neural-style located in pytorch examples repo.
https://github.com/pytorch/examples/blob/master/fast_neural_style/neural_style/neural_style.py