I tried to use the VGG16 model from torchvision.models with the create_cnn() function, and it failed.
Reason I believe is num_features_model() function looks for ‘num_features’ attribute, and none of the layers in the VGG16 model has that attribute.
Using vgg16_bn it works fine, as the batchnorm layers has the ‘num_features’ attribute.
The old VGG16 isn’t supported now, vgg16_bn should work better in any case.