AlexNet & SqueezeNet Comparison on CIFAR10 dataset
It is implemented as Keras on Google Colab (Tesla K80 GPU).
AlexnNet and SqueezeNet is slightly changed to fit in CIFAR10 input image size (32 x 32).
In this repo, SqueezeNet has less Fire layers than original SqueezeNet for comparing AlexNet.
Number of parameters: 22,559,626
Test Accuracy: 79.28%
Number of parameters: 378,914
Test Accuracy: 80.29%
Number of params reduction: ~59x
