Closed zachmayer closed 7 years ago
Might also make sense to use squeezenet as the default: https://github.com/datarobot/imagefeaturizer/issues/4
Then there's now downloading required at all for a quick start.
Done! Added squeezenet with weights packaged into the network, and set it as the default. All models can be called in the ImageFeaturizer() initialization with the model='___' option.
Since we're downloading weights anyways, no reason not to allow the user to choose any of them. Keep the default as one of the inception networks that works on both theanon and tensorflow.