A Py-Torch Implementation of SqueezeNet

SqueezeNet was developed in 2017 by Iandola et al. It achieves similar performance as AlexNet with a much smaller architecture. It exploits ResNet identity block and squeezing operation used by GoogleNet. See their paper. I created this repo because I thought SpueezeNet can serve as a good benchmark for ML techniques, due to its small size and decent performance. I changed the end 13x13 pooling to a couple more fire modules and max pooling with a final 2x2 global pooling. The added layers should allow the model to distill more high level information and indeed helps when training with the Dogs and Cats dataset as well as Caltech 101 or Caltech 256.