How to involve multilabel using VocNet in the model zoo?

[elliot...@gmail.com]

The full name at https://github.com/BVLC/caffe/wiki/Model-Zoo#pascal-voc-2012-multilabel-classification-model is Pascal VOC 2012 Multilabel Classification Model
The models definition in prototxt has a layer which has been commented:

#layer {
#  name: "loss"
#  type: "HingeLossMultiLabel"
#  bottom: "fc8voc"
#  bottom: "label"
#  top: "loss"
#}

This bottom: "label" means it should has a layer has a top named 'label', am I right?
I have seen the this page http://nbviewer.jupyter.org/github/BVLC/caffe/blob/master/examples/pascal-multilabel-with-datalayer.ipynb which introduce how to involve multilabel classification in python, and the model definition:

n.data, n.label = L.Python(module = 'pascal_multilabel_datalayers', layer = datalayer, 
                               ntop = 2, param_str=str(data_layer_params))

has a top named label, and in the 6th step, it uses:
gtlist = test_net.blobs["label"].data[image_index, ...].astype(np.int)
to get the predict label.
1.Then how can I do to let this model output multilabel? Just uncomment the code will get an error about the 'label' thing
2. There is no layer type called HingeLossMultiLabel,does it mean HingeLoss?
Thank you!

[chaip...@gmail.com]

about Q1,you should modify the tool which classification of caffe.

about pascal-voc-2012-multilabel-classification-model,this prototxt is a deploy model,not train model,so it do not have the label for train.

finally,i have same doubt about the type called  HingeLossMultiLabel ,i think the prototxt  of  Pascal VOC 2012 Multilabel Classification Model don't have relationship between the paper " Analyzing classifiers: Fisher vectors and deep neural networks".