VGG net filter questions

[golftlist]

Hi

I am new to CNN, deep learning, Caffe etc and I have some basic questions I hope you can help me. 

I took a look of the trained VGG network, the file VGG_ILSVRC_19_layers.caffemodel is about 600M. I am wondering why it is so big. The first layer of VGG network is 

layers {

  bottom: "data"

  top: "conv1_1"

  name: "conv1_1"

  type: CONVOLUTION

  convolution_param {

    num_output: 64

    pad: 1

    kernel_size: 3

  }

}

Which is 3x3 filter convert 3 color channel to 64 feature channel. My understanding is that for this layers, the total number of weights is 3*3*3*64 =1728. If we use double to save it, it is only 1.3K. Some layers has 512 channel input and 512 channel output and it is about 512*512*3*3=18M. So put them all together, it should be much smaller that 600M. So what else is in the model?

My second question is for 3x3 filer, 3  is half kernel size or full kernel size?

Thanks,