Closed zachmayer closed 7 years ago
Adding squeezenet weights to the package gives us a fallback for cases where it's not possible to download the weights for the other networks.
https://github.com/rcmalli/keras-squeezenet
Different squeezenet: https://github.com/wohlert/keras-squeezenet https://github.com/wohlert/keras-squeezenet/blob/master/squeezenet_weights.h5
Done! Added rcmalli's version and set SqueezeNet as the default network, loaded from weights. It obtains 94% accuracy on the Cats vs. Dogs dataset after just slapping a linear classifier on top.
Adding squeezenet weights to the package gives us a fallback for cases where it's not possible to download the weights for the other networks.
https://github.com/rcmalli/keras-squeezenet