this code implement VGG16 network by this paper.
- load pre-trained parameters from here: https://mega.nz/#!YU1FWJrA!O1ywiCS2IiOlUCtCpI6HTJOMrneN-Qdv3ywQP5poecM
- the pre-trained parameters were trained on ImageNet DataSet, 1000 clasess.
- Remove the final FC layer and add one 10 nodes FC layer to apply for Cifar10 DataSet https://www.cs.toronto.edu/~kriz/cifar.html
use the pre_trained convolution layer parameters, and train the FC layer parameters
Do not use the pre-trained parameter, the network train all parameters