Performance on ImageNet data (and similarly sized images)

[seansaito]

I notice that the model by default assumes images of dimensions 64x64. I'm curious how the model/distance metric performs for higher resolutions like that of ImageNet (224x224).

[richzhang]

The model is fully-convolutional, so it can be run at any resolution. All our empirical tests are on 64x64 though.

[seansaito]

Thanks for the reply! I see, any intuitions on how well it would work on 224x224 data?

[richzhang]

The edge effects will be slightly different, but it should work okay.