How to predict VOC multi-label classification?

[刘宇]

H, friends,

I plan to do some experiments about VOC classification. 

But each image has several labels, so how can I prepare the Data and Labels (I mean the convert_imagenet.bin)?

For example:

XXX.jpg 1 1 0 ...

Is that right?

Moreover, which loss function and accuracy layer is fit for multi-label task?

Thousand thanks in advance!!