Face recognition from scratch - GoogLeNet, AlexNet or VGG?

[Thomas Wood]

Hi

I am trying to develop my own face recogniser to run on mobile platforms. I’m using Caffe via DIGITS.

I have my own crawled training data, but I’m not sure what’s the best network configuration. It should do the same as VGG Face Descriptor but I can’t use that because of the licence

1. GoogLeNet - either from scratch or fine tuned
2. AlexNet - either from scratch or fine tuned
3. My own network similar to VGG Face Descriptor
4. My own network approx similar to GoogLeNet/AlexNet

Since DIGITS offers GoogLeNet out of the box, I was wondering, should I just take GoogLeNet and supply it with my crawled images. Or is there a reason to use a different network?

Does the VGG face descriptor offer any advantages over GoogLeNet for face recognition? I can see that both have a lot of layers but I am not sure if there is anything about the VGG configuration that's specific to face. What are the relative advantages between GoogLeNet/AlexNet? I can see GoogLeNet is smaller than AlexNet.

Bearing in mind I need this to be a small and fast network to run on mobile, I am veering towards GoogLeNet but I'm interested to hear any suggestions.

Thanks

Tom