Adding parameters with github lfs

[317070]

I have added the parameters into a Lasagne readable format for vgg-16, vgg-19 and googlenet. These parameters are stored on git large file storage: https://git-lfs.github.com/ The result of these networks are equal except for floating point rounding with the original caffe models.

[ebenolson]

Thanks, but as I mentioned in https://github.com/Lasagne/Recipes/issues/4, I don't think Github LFS is a great fit for this - their free tier quota is quite limited and already exceeded by this PR...

Since these model definitions have been available for a while and are used/copied in several of the example notebooks, I'd rather not change them without a strong reason. I also personally prefer the compactness of the current format, and would like to avoid a dependency on pycaffe.

[f0k]

I'd rather not change them without a strong reason.

+1, there might even be users relying on the current interface (i.e., being able to import the file and calling build_model()). If we're changing the files, the interface should remain compatible.

I also personally prefer the compactness of the current format,

+1! Yours are a lot easier to read.

and would like to avoid a dependency on pycaffe.

+1. Note that https://github.com/Lasagne/Recipes/blob/master/examples/Using%20a%20Caffe%20Pretrained%20Network%20-%20CIFAR10.ipynb already shows how to get a Caffe model into Lasagne, so there's no reason to replicate it in modelzoo/googlenet.py. It would be nice to have a general converter script in modelzoo/, though, as that's where people would look for it.