Question about siamese network example.

[Jun-ya Norimatsu]

Hi,

I'm trying to use caffe for siamese network.
I checked the code of examples/siamese/convert_mnist_siamese_data.cpp and have a question about the generating strategy of the training data.

The question is:
It seems that the positive and negative data generated may be unbalanced. (I mean that the amounts of positive data and negative data are not the same.)
Is there any reason for the strategy?

I wonder why it works well.

Thanks in advance,
--
Jun-ya NORIMATSU