I want a dump of the last POOL layer of the VGG16, for the ImageNet database (or its part).
In other words, you would need to take an image from ImageNet, feed it into a pre-trained VGG16 network, and write down the values in the last pooling layer.
Probably, you can use Google Colab, with torch / pytorch, with pretrained vgg16 model already preinstalled. I think, that ImageNet database is available within pytorch, as well. I learned this myself some time ago ;) but I forget quickly.